CHALLENGING THE FOUNDATION OF COMPUTER SCIENCE ...

2015-16 ANNUAL REPORT

TO Y O TA T E C H N O L O G I C A L I N S T I T U T E AT C H I C A G O

TO Y O TA T E C H N O L O G I C A L I N S T I T U T E AT C H I C A G O

CONTENTS Institute Mission Message from the President Note from the Chief Academic Officer Institute Overview Awards and Honors

4

New Faculty Faculty Promotion and Tenure Faculty by Area Postdocs Research and Responsibility

11

6 8 9 10 12 13 13 14

Research Philosophy Algorithms and Complexity Computational Biology Computer Vision and Computational Photography Machine Learning Robotics Speech and Language Technologies

Collaboration and Cooperation Talks, Seminars and Workshops Education The PhD Program Student Progress TTIC Curriculum Servicing the University of Chicago TTIC Student Awarded University of Chicago TA Prize Student Publications, Posters, Abstracts TTIC’s Largest Incoming Class Financial Support Exchange Students

2

40 42 50

Institute Goals

56

Accreditation Board Strategy for Continued Improvements External Advisory Committee Visit 2016 Endowment Growth and Investment Collaboration with University of Chicago Computing Capacity Doubles in 2016

Interns and Visiting Scholars

60

Institute Financial Reports Faculty Research Highlights

62

Governance

70

64

Board of Trustees Leadership Administration

Equal Opportunity Statement

73

Special Thanks

74

3

INSTITUTE MISSION Achieving international impact through world-class research and education in fundamental computer science and information technology.

The Research Mission TTIC aims to achieve international impact through world-class research in fundamental computer science and information technology. Here, we clarify the intended meaning of the terms in this statement. Impact. The mission statement focuses on academic impact. A number of criteria may serve to evaluate such impact. These include volume of peer-reviewed publications; reputation of venues in which publications appear; visibility of work in the community, as expressed in citations by others; number and reputation of co-authors, in particular in other institutions; recognition by research community, including awards, prizes, invited talks, and invitation or election to serve in senior service positions in professional organizations; reports by external advisory bodies comprised of reputable senior researchers, etc. Precise objective measures of academic impact are controversial and elusive, and no one of the criteria above is alone a solid measure in itself. However, the combined evaluation of these and similar criteria helps assess the academic impact achieved by TTIC researchers. Note that the number of patents filed, or the amount of extramural research funding, are not considered measures of academic impact. Although funding is clearly an important tool in achieving impact, it is only a tool and not an end in itself. Fundamental. The mission statement is intended to focus on scientifically fundamental research. A scientific result is fundamental to the extent that it has open-ended implications. It is important to distinguish being fundamental from being economically important. A calendar program can be economically successful, and hence important, without adding to fundamental knowledge. The concept of NP-completeness adds greatly to the fundamental understanding of computation without having clear economic significance.

4

Computer Science and Information Technology. Computer science and information technology encompasses many sub-disciplines. In the selection of sub-disciplines for study at TTIC there should be some consideration of relevance to society as a whole. The interpretation of “computer science” and “information technology” should be such that TTIC remains relevant to the societal impact of computation and information.

The Education Mission The educational mission of TTIC is to achieve international impact through the accomplishments of its graduates as productive scientists and citizens. The notion of “impact” in the educational mission is broader than in the research mission. The graduates of TTIC might achieve impact by starting successful companies, managing successful products, or influencing government directions in research funding. Of course, TTIC also strives to produce PhDs who achieve academic impact throughout their careers. The institute strives to produce graduates who contribute to society through their intellectual leadership in computer science and information technology. Success in the educational mission requires appropriate selection of curriculum, effective teaching to enable learning, effective assessment and mentorship of students, and effective marketing of students in the job market. TTIC strives to place its PhD graduates at high quality research institutions. TTIC also strives to make its PhD students visible to the academic community before graduation. This can be done most effectively through publications prior to graduation.

5

MESSAGE FROM THE PRESIDENT During 2015-2016, the Toyota Technological Institute at Chicago (TTIC) continued its steady progress towards fulfilling its mission: achieve international impact through world-class research and education in fundamental computer science and information technology. Four new non-tenure-track faculty members will arrive and four new students will enroll at the start of the 2016-17 academic year. In recognition of his outstanding research, professional stature, and contributions to TTIC, Dr. Greg Shakhnarovich has been promoted to Associate Professor with tenure. The level of recognition and impact of research at TTIC continues to steadily increase, the latest examples being the Best Paper Award given to Li-Yang Tan at the 56th Annual Symposium on Foundations of Computer Science (FOCS 2015) for the paper "An Average-Case Depth Hierarchy Theorem for Boolean Circuits," co-authored with Benjamin Rossman and Rocco Servedio, and the Best Student Paper Award given to Hao Tang at the IEEE ICASSP conference for “Signer-independent fingerspelling recognition with deep neural network adaptation,” The faculty actively pursued federal research grants with sufficient success that the grant volume kept about the same amount as the last year. 2015-16 was a significant year for the development of TTIC’s educational program and institutional structure. Under the accreditation renewal process, TTIC received formal notification from the Institutional Action Council (IAC) of the Higher Learning Commission of the North Central Association of Colleges and Schools (HLC) in March 2015, stating some concerns and informing us that TTIC was in danger of being placed on Notice, the first level of sanction. The concerns involved: (a) its financial independence, (b) autonomy of the Board of Trustees to make decisions for the future of the institution, (c) assessment of study learning, and (d) retention, persistence, and completion rate. TTIC and its Board have made significant progress and have implemented thoughtful changes to address the concerns. In July 2015, TTIC submitted its report to the HLC, and at an Institutional Actions Council hearing held on

6

September 1, 2015, our team explained what actions the institute had undertaken in the nine months prior to the hearing, engaged in Q&A with the panel, and promoted their understanding of TTIC’s improvement strategies. On September 11, 2015, TTIC received the official action letter and report from the HLC. The letter did not recommend sanction but instead recommended accreditation renewal for ten years with an interim report that addresses TTIC progress on HLC concerns by 2017, and an additional site visit in four years. We believe that this process has strengthened TTIC in a variety of dimensions. The HLC letter also requested TTIC make a Substantive Change Application to ask for accreditation of the institute’s Master’s within the PhD program degree. TTIC completed the application, and maintained all documentation and preparation to accept an HLC team visit on June 20 and 21, 2016. The HLC team visit was successful, and they made an official report recommending to approve the application. On August 19, TTIC received an official letter from the HLC’s IAC stating that they approved the TTIC’s

application and

the outcome would be publicly posted online in early September. Our relationship with Toyota Technological Institute (TTIJ) in Nagoya continues to strengthen. Two TTIJ exchange students spent a quarter at TTIC during 2015-2016 academic year, and several TTIJ faculty spent time at TTIC conducting joint research with our faculty. The institute’s relationship with the University of Chicago remains strong, both in respect to various kinds of administrative aid, student support and to the potential for collaborative research and academic endeavors. TTIC is recently involved in collaboration agreements with several Japanese institutes, and newly established an agreement with the Artificial Intelligence Research Center of the National Institute of Advanced Industrial Science and Technology. We look forward to fruitful collaboration with many more research institutes. As TTIC continues to mature as an institution, we are committed in furthering improvement of academic excellence and to enhancing the already strong relations with our academic partners. We will continue hiring the strongest faculty possible.

Sadaoki Furui President

7

NOTE FROM THE CHIEF ACADEMIC OFFICER 2015-16 was a fairly eventful year for TTIC. We completed an accreditation renewal. We held the first Midwest Robotics Workshop and the annual Midwest Vision Workshop. We held a successful distinguished lecture series. Research Assistant Professor Li-Yan Tang won a very prestigious Best Paper Award at FOCS 2015, one of the two flagship conferences in theoretical computer science. PhD candidate Hao Tang won a Best Student Pnaper award at ICASSP 2016. Greg Shakhnarovich was promoted to Associate Professor with Tenure. Congratulations, Greg! This year we were very sorry to say goodbye to Anna Ruffolo who served TTIC for the better part of a decade as Controller and Director of Operations. Anna left TTIC for a position as Senior Vice President of Finance and Operations at a considerably larger organization, an offer she could not refuse. We all have fond memories of Anna and were very sorry to see her go. In May 2016, we welcomed Jessica Johnston as Chief Financial Officer. Jessica was most recently Controller at the American College of Chest Physicians. Welcome, Jessica! The world around us continues to change rapidly. This year saw continued dramatic improvements in computer vision systems due to the continuing evolution of deep neural network methods. The error rate on the flagship benchmarks in computer vision was roughly cut in half from the previous year. In another development, the computer system Alphago defeated Li Sedol, one of, if not the, best Go player in the world. This level of computer play in Go was not expected for decades. Research in self-driving cars continued to accelerate, with the number of companies involved growing to over thirty. Many of these companies have announced plans to have products on the market within five years. Industry continues to play an increasing role in basic research in artificial intelligence and deep learning. Sakichi Toyoda famously said, “Respect the spirit of research and creativity, and always strive to stay ahead of the times.” Staying ahead of the current times appears to require some serious striving. We will strive.

David McAllester Chief Academic Officer

8

INSTITUTE OVERVIEW Faculty and Staff Professors

1

Associate Professors

4

Assistant Professors

5

Research Assistant Professors

13

Adjoint Faculty

7

Administrative Office Staff and IT

8

PhD Program Students Enrolled for 2015-16

28

Master’s within the PhD Program Degrees Awarded

4

PhD Degree Requirement Completion

1

Applicants for the 2015-16 Academic Year

89

Admitted

21

Enrolling

9

Exchange Students in 2015-16

2

9

AWARDS AND HONORS 2016 March

Hao Tang

PhD Candidate Hao Tang received Best Student Paper Award at the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2016) held in Shanghai, China for “Signerindependent fingerspelling recognition with deep neural network adaptation,” co-authored by Taehwan Kim, Weiran Wang and Karen Livescu.

2016 March

Jian Peng

TTIC alumnus Jian Peng has been named a 2016 Sloan Research Fellow by the Alfred P. Sloan Foundation. Jian Peng received his bachelor's and master's degrees in Computer Science at Wuhan University, and his PhD in Computer Science from TTIC in 2013. Peng went on to postdoctoral research at the Berger Lab of Massachusetts Institute of Technology and is currently an assistant professor in the Computer Science Department at the University of Illinois, Urbana-Champaign. Peng was also the recipient of a Microsoft PhD Fellowship while studying at TTIC. The Alfred P. Sloan Research Fellowship was founded in 1955 to encourage young scholars with great achievements or huge potential in the fields of physics, math and chemistry. Fields have since expanded to include neuroscience, economics, computer science and computational and evolutionary molecular biology. Each Fellow is awarded $50,000 to support their early career development. Since the beginning of the program in 1955, forty-three fellows have won the Nobel Prize in their respective field, sixteen have received the Fields Medal in mathematics, and many have become eminent talents of their fields.

2015 October

Li-Yang Tan

Prof. Li-Yang Tan was awarded Best Paper Award for The IEEE Symposium on Foundations of Computer Science (FOCS 2015) held in Berkeley, CA, for the paper, “An average-case depth hierarchy theorem for Boolean circuits,” with co-authors Benjamin Rossman and Rocco A. Servedio.

10

NEW FACULTY Srinadh Bhojanapalli

Kevin Gimpel

Research Assistant Professor PhD - University of Texas, Austin

Assistant Professor PhD - Carnegie Mellon University

Srinadh Bhojanapalli obtained his M.S. and PhD in Electrical and Computer Engineering from The University of Texas at Austin in 2012 and 2015, respectively. Prior to that he obtained Bachelors in Technology from Indian Institute of Technology Bombay in 2010. He has spent some summers as intern at Microsoft Research India and ebay Research labs. He is currently a Research Assistant Professor at TTIC.

Kevin Gimpel received a B.S.E. in Computer Science from the University of Pennsylvania in 2004 and a PhD from the Language Technologies Institute at Carnegie Mellon University in 2012. He also worked at MIT Lincoln Laboratory from 2004 to 2006 and interned with the machine translation team at Google in 2009.

His research is primarily focused on designing algorithms for large scale machine learning problems with statistical guarantees. He is interested in Matrix and Tensor factorization, Optimization and Sublinear time algorithms. Recently he has been working on designing scalable algorithms for Semidefinite Optimization with provable convergence guarantees.

His research interests include several areas of computational linguistics, focusing on machine translation, statistical natural language processing, and machine learning for models of language. He is also interested in social media analysis, computational semantics, and forecasting real-world events using text data. Dr. Gimpel also has a personally maintained website which can be found at www.ttic.edu/ gimpel.

Dr. Bhojanapalli also has a personally maintained website which can be found at www.ttic.edu/ bhojanapalli.

Li-Yang Tan Research Assistant Professor PhD - Columbia University Li-Yang Tan received his PhD in 2014 from Columbia University, where he was advised by Rocco Servedio. He spent the 2014-15 academic year as a Microsoft Research Fellow at the Simons Institute at University of California, Berkeley, before joining TTIC in June 2015. His research interests lie in theoretical computer science, with an emphasis on computational complexity. Specific interests include concrete complexity, the analysis of Boolean functions, property testing, and computational learning theory. Dr. Tan also has a personally maintained website which can be found at www.ttic.edu/tan.

11

FACULTY PROMOTION AND TENURE At the April 2016 meeting of the Board of Trustees, upon recommendation of the President, the Trustees approved Professor Greg Shakhnarovich for promotion to Associate Professor, with tenure.

Prof. Shakhnarovich began his appointment with TTIC in autumn of 2007. He teaches the course TTIC 31020: Introduction to Statistical Machine Learning, a core course in the PhD curriculum, and TTIC’s highest attended course by both TTIC and University of Chicago students. Prof. Shakhnarovich also serves as faculty IT liaison, working with the institute Director of IT to represent academic needs and planning, and as Admissions Director, working with senior faculty and institute administration to strategize recruitment efforts and execute the institute admissions goals for the year. Prof. Shakhnarovich is advising six students at TTIC: three in Candidacy, and three in the early stages of the PhD program. He played an active role in assisting with the ongoing accreditation process at TTIC, and contributed to the improved processes that were outcomes of the exercise. He is also the main organizer for the 2016 Midwest Vision Workshop held at the institute in the spring. Prof. Shakhnarovich has over thirty refereed publications.

12

FACULTY BY AREA Algorithms & Complexity

Machine Learning

Chuzhoy, Julia

Bhojanapalli, Srinadh

Fortnow, Lance

Garber, Dan

Li, Shi

Lafferty, John

Makarychev, Yury

Mahdavi, Mehrdad

Nguyen, Huy Le

McAllester, David

Razborov, Alexander

Meshi, Ofer

Tan, Li-Yang

Mita, Seichi

Tulsiani, Madhur

Nowak, Robert

Wright, Stephen

Sasaki, Yutaka Srebro, Nathan Tomioka, Ryota

Computational Biology Canzar, Stefan Huang, Qixing Khan, Aly Naveed, Hammad

Robotics Walter, Matthew

Xu, Jinbo

Computer Vision & Computational Photography

Speech and Language Technologies

Chakrabarti, Ayan

Bansal, Mohit

Forsyth, David

Gimpel, Kevin

Maire, Michael

Livescu, Karen

Shakhnarovich, Greg

Roth, Dan

Post Doc Wang, Weiran

13

RESEARCH & RESPONSIBILITY Research Philosophy Research is the heart and soul of activity at the Toyota Technological Institute at Chicago. The institute has an energetic and determined team of professors, visiting professors, assistant professors, research assistant professors, adjoint professors and postdocs encompassing many areas of research interests, and from many countries, backgrounds, each bringing their own specialty to the Institute.

With a generous budget, distinguished professors, and an environment that promotes learning and sharing, there are ample opportunities for collaborative research. Being on the campus of the University of Chicago, there is opportunity for close and cooperative research with not only the University of Chicago Computer Science Department, but with the departments of Mathematics, Statistics, and most recently, the Booth Graduate School of Business. There are also many guests and visitors who come to TTIC to give talks, participate in workshops, and share their research findings, all heightening the feeling of enthusiasm that pulses through the Institute.

The mission of TTIC includes “…achieving international impact through world-class research and education in fundamental computer science and information technology.” The research component of the mission is implemented through high quality research in high impact areas. Currently, there are active research programs in six areas: machine learning, algorithms and complexity, computer vision and computational photography, speech and language technologies, computational biology, and robotics. The areas are introduced below, and in some, TTIC’s strategy for achieving impact is also described. A key part of the strategy for achieving impact in all areas is to foster collaboration and communication between the areas.

14

Algorithms and Complexity One of the central tasks in all areas of computer science is the writing of efficient software to perform required computation. In order to write such software, one must first design an efficient algorithm for the computational task at hand. The area of algorithms focuses on designing algorithms, and more generally developing powerful algorithmic tools, for solving fundamental computational problems that frequently occur in different areas of computer science. Complexity theory is the study of the power and the limits of efficient computation. The central problem studied by complexity theorists is “Which computational problems can, and which cannot, be solved efficiently?” The study of algorithms and complexity is a part of a broader area called “theory of computer science,” or just “theory.” The area of theory works on developing theoretical foundations for computer science, which lead to a deeper understanding of computation in general, and specific computational tasks in particular, which include better algorithms and faster software. Below is a list of the work done at TTIC this year in the area of Algorithms and Complexity.

Julia Chuzhoy Associate Professor www.ttic.edu/chuzhoy PUBLISHED/SUBMITTED PAPERS Chuzhoy, Julia, and David H.K. Kim. “On Approximating Node-Disjoint Paths in Grids.” Paper presented at the International Workshop on Approximation Algorithms for Combinatorial Optimization Problems (APPROX), Princeton University, August 2015. doi:10.4230/ LIPIcs.APPROX-RANDOM.2015.187. “Large-Treewidth Graph Decompositions.” In Encyclopedia of Algorithms. Springer New York, 2016. doi:10.1007/978-1-4939-2864-4_691 Chuzhoy, Julia, Yury Makarychev, Aravindan Vijayaraghavan, Yuan Zhou. “Approximation Algorithms and Hardness of the k-Route Cut Problem.” Special issue, ACM Transactions on Algorithms 12, no. 1 (February 2016). doi:10.1145/2644814. Chuzhoy, Julia, David H.K. Kim, and Shi Li. “Improved Approximation Algorithms for NodeDisjoint Paths in Planar Graphs.” Paper presented at the ACM Symposium on Theory of Computing (STOC), Cambridge, June 2016. doi:10.1145/2897518.2897538. Chuzhoy, Julia, and Alina R. Ene. “On Approximating Maximum Independent Set of Rectangles.” Paper presented at the Annual IEEE Symposium on Foundations of Computer Science (FOCS), New Brunswick, October 2016. arXiv:1608.00271. TALKS “Excluded Grid Theorem: Improved and Simplified.” Plenary talk given at the 7th Workshop on Graph Optimization and Width Parameters, Aussois, France, October 2015. “Excluded Grid Theorem: Improved and (somewhat) Simplified.” Plenary talk given at Graph Theory Workshop, Oberwolfach, Germany, January 2016. “Excluded Grid Theorem: Improved and Simplified.” Talk given at Harvard/MIT/MSR Theory Seminar, Cambridge, MA, February 2016. “Approximation Algorithms for Graph Routing Problems.” Colloquium Distinguished Speaker talk given at Max Planck Institute at Saarbrucken, Germany, June 2016. “Excluded Grid Theorem: Improved and Simplified.” Invited talk given at 15th Scandinavian Symposium and Workshops on Algorithm Theory, Reykjavik, Iceland, June 2016. “Excluded Grid Theorem: Improved and Simplified.” Plenary talk given at 2016 SIAM Meeting on Discrete Mathematics, Atlanta GA, June 2016.

15

INVOLVEMENT Editorial Board Member, SIAM Journal on Computing Editor, FOCS 2014 special issue, SIAM Journal on Computing Co-organizer, “Highlights of Algorithms” Workshop, Paris, June 2016 Reviewer: FOCS 2016, ICALP 2016, SODA 2016 Co-organizer, “Approximation Algorithms and the Hardness of Approximation” BIRS Workshop, Alberta, Canada, to be held November 2017 RESEARCH FUNDING AWARDS NSF Grant, “AF: Small: Graph Routing, Vertex Sparsifiers, and Connections to Graph Theory.” September 2016-August 2019: $449,720 CLASSES/SEMINARS TTIC 31080 - Approximation Algorithms (CMSC 37503): The course covered several advanced topics related to the area of Approximation Algorithms, including: Graph Minor Theory; routing problems; iterative rounding technique; Lovász Local Lemma; oblivious routing and related graph decompositions. MISCELLANEOUS Organizer, TTIC Distinguished Lecture Series Member, Search Committee for CS Department Chair, University of Chicago Chair, Search Committee for Chief Academic Officer, TTIC Student Support Coordinator Minute Taker and Facilitator, TTIC Faculty meetings Student supervision: Rachit Nimavat (TTIC) David H.K. Kim (UChicago CS department; co-supervised with Laszlo Babai) Joshua Kaplan (UChicago undergraduate student, summer internship in summer 2015 and REU during the 2015/2016 academic year) Vivek Madan (UIUC, summer intern, summer 2016) Sepideh Mahabadi (MIT, summer intern, summer 2016)

Yury Makarychev Assistant Professor www.ttic.edu/makarychev PUBLISHED/SUBMITTED PAPERS Makarychev, Konstantin, Yury Makarychev, and Yuan Zhou. “Satisfiability of Ordering CSPs Above Average Is Fixed-Parameter Tractable.” Paper presented at the IEEE Symposium on Foundations of Computer Science (FOCS), Berkeley, October 2015. doi:10.1109/ FOCS.2015.64. Makarychev, Konstantin, and Yury Makarychev. “Union of Euclidean Metric Spaces is Euclidean.” Discrete Analysis 14 (2016). doi:10.19086/da.876. Makarychev, Konstantin, Yury Makarychev, and Aravindan Vijayaraghavan. “Learning Communities in the Presence of Errors.” Paper presented at Conference on Learning Theory (COLT), New York City, June 2016. arXiv:1511.03229v3. Makarychev, Konstantin, Yury Makarychev, Maxim Sviridenko, and Justin Ward. “A Bi-criteria Approximation Algorithm for k-Means.” Paper to be presented at the International Workshop on Approximation Algorithms for Combinatorial Optimization Problems (APPROX), Paris, September 2016. arXiv:1507.04227v2. Makarychev, Konstantin, and Yury Makarychev. “Bilu–Linial Stability.” In Advanced Structured Prediction, ed. Tamir Hazan, George Papandreou, and Daniel Tarlow. Cambridge: MIT Press, forthcoming.

16

TALKS “Satisfiability of Ordering CSPs Above Average Is Fixed-Parameter Tractable.” Workshop on Constraint Satisfaction Problems, Schloss Dagstuhl, Germany, July 2015. “Satisfiability of Ordering CSPs Above Average Is Fixed-Parameter Tractable.” Talk given at Microsoft Research, Redmond, WA, October 2015. “ Non-uniform Graph Partitioning with Unrelated Weights.” Workshop on Approximation Algorithms, Institute for Mathematical Sciences, National University of Singapore, February 2016. “The Grothendieck Constant is Strictly Smaller than Krivine's Bound.” Colloquium talk given at the Department of Applied Mathematics, Illinois Institute of Technology, Chicago, February 2016. “Metric Techniques in Computer Science.” Lecture series given at concentration week in Metric Spaces: Analysis, Embeddings into Banach Spaces, Applications; Texas A&M, College Station, TX, July 2016 (jointly with K. Makarychev). INVOLVEMENT Conference Reviews: STOC 2016, FOCS 2016, SODA 2016, COLT 2016, ICALP 2016, STACS 2016 Reviewer, Theory of Computing Reviewed grant proposals for the Israel Science Foundation and Swiss National Science Foundation Programming Experience Czar, TTIC RESEARCH FUNDING AWARDS NSF Career Award CCF-1150062. NSF Grant IIS-1302662 (jointly with N. Srebro). CLASSES/SEMINARS TTIC 31010 / CMSC 37000-1 (Algorithms): This is a graduate level course on algorithms with the emphasis on central combinatorial optimization problems and advanced methods for algorithm design and analysis. Topics covered include asymptotic analysis, greedy algorithms, dynamic programming, amortized analysis, randomized algorithms and probabilistic methods, combinatorial optimization and approximation algorithms, linear programming, and advanced data structures.

Huy Le Nguyên Research Assistant Professor www.ttic.edu/nguyen PUBLISHED/SUBMITTED PAPERS Kamma, Lior, Robert Krauthgamer, and Huy L. Nguyen. “Cutting Corners Cheaply, or How to Remove Steiner Points.” SIAM Journal on Computing 44, no. 4 (2015): doi:10.1137/140951382. Barbosa, Rafael P., Alina Ene, Huy L. Nguyen, and Justin Ward. “The Power of Randomization: Distributed Submodular Maximization on Massive Datasets.” Paper presented at the International Conference on Machine Learning (ICML), Lille, France, July 2015. Ene, Alina, and Huy L. Nguyen. “Random Coordinate Descent Methods for Minimizing Decomposable Submodular Functions.” Paper presented at the International Conference on Machine Learning (ICML), Lille, France, July 2015. Andoni, Alexandr, and Huy L. Nguyen. “Width of Points in the Streaming Model.” ACM Transactions on Algorithms 12, no. 1 (February 2016): doi:10.1145/2847259. Braverman, Mark, Ankit Garg, Tengyu Ma, Huy L. Nguyen, and David P. Woodruff. “Communication Lower Bounds for Statistical Estimation Problems via a Distributed Data

17

Processing Inequality.” Paper presented at the Annual Symposium on the Theory of Computing (STOC), Cambridge, June 2016. Larsen, Kasper G., Jelani Nelson, Huy L. Nguyen, and Mikkel Thorup. “Heavy Hitters via Cluster -Preserving Clustering.” Paper to be presented at the IEEE Symposium on Foundations of Computer Science (FOCS), New Brunswick, October 2016. Ene, Alina, and Huy L. Nguyen. “Constrained Submodular Maximization: Beyond 1/e.” Paper to be presented at the IEEE Symposium on Foundations of Computer Science (FOCS), New Brunswick, October 2016. Barbosa, Rafael P., Alina Ene, Huy L. Nguyen, and Justin Ward. “A New Framework for Distributed Submodular Maximization.” Paper to be presented at the IEEE Symposium on Foundations of Computer Science (FOCS), New Brunswick, October 2016. TALKS “Distributed Machine Learning.” Talk given at Departmental Colloquium, Department of Computer Science, University of Iowa, September 2015. “Distributed Submodular Maximization in MapReduce.” Talk given at Sublinear Algorithms Workshop, John Hopkins University, Baltimore, January 2016. “Communication Lower Bounds for Statistical Estimation Problems via a Distributed Data Processing Inequality.” Talk given at Information Theory Reunion Workshop, Simons Institute, UC Berkeley, June 2016. INVOLVEMENT Program Committee Member, COCOON 2016 Conference Reviews: NIPS 2015; SODA 2016; STOC 2016; ICML 2016; CCC 2016; FOCS 2016; NIPS 2016 Journal Review: SIAM Journal on Computing (SICOMP)

Razborov, Alexander Adjoint Professor www.ttic.edu/razborov PUBLISHED/SUBMITTED PAPERS “A New Kind of Tradeoffs in Propositional Proof Complexity.” Journal of the ACM 62, no. 3 (May 2016). doi:10.1145/2858790. Coregliano, Leonardo N. and Alexander A. Razborov, “On the Density of Transitive Tournaments.” Journal of Graph Theory (May 2016). doi:10.1002/jgt.22044. TALKS “Continuous Combinatorics.” Oberwolfach workshop, Germany, 2015. “Continuous Combinatorics.” Talk given at Chebyshev Lab Colloquium, St. Petersburg State University, Washington, June 2015. “Continuous Combinatorics.” Talk given at Microsoft Research Redmond, May 2016. “On the AC0 Complexity of Subgraph Isomorphism.” Talk given at the Theory of Computation Seminar, Harvard University, Cambridge; November 2015. “Complexity of Semi-Algebraic and Algebraic Proofs.” Talk given at the Trends in Optimization Seminar, University of Washington, Seattle; Carnegie-Mellon University; TTIC; Moscow; and St. Petersburg State University, May 2016. INVOLVEMENT Editorial boards: Forum of Mathematics, Pi and Sigma; Izvestiya of the Russian Academy of Science, ser. mathem.; Combinatorica; Combinatronics, Probability and Computing

18

CLASSES/SEMINARS CMSC 27100 - Discrete Mathematics: University of Chicago: This course emphasizes mathematical discovery and rigorous proof, which are illustrated on a refreshing variety of accessible and useful topics. Basic counting is a recurring theme and provides the most important source for sequences, which is another recurring theme. Further topics include proof by induction; recurrences and Fibonacci numbers; graph theory and trees; number theory, congruences, and Fermat's little theorem; counting, factorials, and binomial coefficients; combinatorial probability; random variables, expected value, and variance; and limits of sequences, asymptotic equality, and rates of growth. CMSC 38410 - Quantum Computing: In this course we will discuss mathematical and complexity aspects of quantum computing, putting aside all questions pertaining to its physical realizability. CMSC 37120 - Topics in Discrete Mathematics: Arithmetic Combinatorics

Li-Yang Tan Research Assistant Professor www.ttic.edu/tan PUBLISHED/SUBMITTED PAPERS Blais, Eric, Clément L. Canonne, Igor C. Oliveira, Rocco A. Servedio, and Li-Yang Tan. “Learning Circuits with Few Negations.” Paper presented at the International Workshop on Randomization and Computation (RANDOM), Princeton University, August 2015. doi: 10.4230/LIPIcs.APPROX-RANDOM.2015.512. Dughmi, Shaddin, Nicole Immorlica, Ryan O’Donnell, and Li-Yang Tan. “Algorithmic Signaling of Features in Auction Design.” Paper presented at the International Symposium on Algorithmic Game Theory (SAGT), Saarbrücken, Germany, September 2015. doi:10.1007/978-3-66248433-3_12. Rossman, Benjamin, Rocca A. Servedio, and Li-Yang Tan. “An Average-Case Depth Hierarchy Theorem for Boolean Circuits.” Paper presented at the IEEE Symposium on Foundations of Computer Science (FOCS), Berkeley, October 2015. doi:10.1109/FOCS.2015.67. Kauers, Manuel, Ryan O'Donnell, Li-Yang Tan, and Yuan Zhou. “Hypercontractive Inequalities via SOS, and the Frankl-Rodl Graph.” Discrete Analysis, February 2016. doi:10.19086/ da.612. Chen, Xi, Igor C. Oliveira, Rocco A. Servedio, and Li-Yang Tan. “Near-Optimal Small-Depth Lower Bounds for Small-Distance Connectivity.” Paper presented at the Annual ACM SIGACT Symposium on Theory of Computing (STOC), Cambridge, June 2016. doi:10.1145/2897518.2897534. Pitassi, Toniann, Benjamin Rossman, Rocco A. Servedio, and Li-Yang Tan. “Poly-logarithmic Frege Depth Lower Bounds via an Expander Switching Lemma.” Paper presented at the Annual ACM SIGACT Symposium on Theory of Computing (STOC), Cambridge, June 2016. doi:10.1145/2897518.2897637. TALKS “An Average-Case Depth Hierarchy Theorem for Boolean Circuits.” Talk given for TCS+ Online Seminar series, October 2015. “An Average-Case Depth Hierarchy Theorem for Boolean Circuits + Circuit Complexity of Small Distance Connectivity.” Talk given at Theory of Computing Seminar, KTH Royal Institute of Technology, Stockholm, October 2015. “An Average-Case Depth Hierarchy Theorem for Boolean Circuits.” Talk given at Oberwolfach Computational Complexity Workshop, November 2015. “An Average-Case Depth Hierarchy Theorem for Boolean Circuits.” Talk given at Spectral Graph Theory Reunion Workshop, UC Berkeley Simons Institute, December 2015. “An Average-Case Depth Hierarchy Theorem for Boolean Circuits.” Theory of Computing Seminar, Harvard University, April 2016.

19

“An Average-Case Depth Hierarchy Theorem for Boolean Circuits II.” Talk given at Computer Science and Discrete Math Seminar, Institute for Advanced Study, Princeton, NJ, April 2016. “An Average-Case Depth Hierarchy Theorem for Boolean Circuits.” Theory of Computing Seminar, University of Chicago, May 2016. “Unconditional Lower Bounds in Complexity Theory.” Talk given at Theory of Computation Colloquium, Massachusetts Institute of Technology, December 2015. “Circuit Lower Bounds via Random Projections.” Talk given at Theory of Computing Seminar, Columbia University, New York, February 2016. HONORS/AWARDS Best Paper Award (with Benjamin Rossman and Rocca Servedio), FOCS 2015 RESEARCH FUNDING AWARDS NSF Algorithmic Foundations Award 1563122, “Circuit Lower Bounds via Random Projections.” Medium, Collaborative Research (with Rocca Servedio), 2016-2020: $1,190,000 ($356,000 awarded to TTIC). MISCELLANEOUS Rossman, Benjamin, and Li-Yang Tan. “Research Vignette: Hard Problems All the Way Up.” newsletter, August 18, 2015. https://simons.berkeley.edu/news/research-vignette-Tan-Rossman -2015.

Madhur Tulsiani Assistant Professor and Director of Graduate Studies www.ttic.edu/tulsiani PUBLISHED/SUBMITTED PAPERS Hatami, Pooya, Sushant Sachdeva, and Madhur Tulsiani. “An Arithmetic Analogue of Fox's Triangle Removal Argument.” Online Journal of Analytic Combinatorics, no. 11 (2016). arXiv:1304.4921v3. Syred, Ridwan, and Madhur Tulsiani. “Proving (Weak) Approximability without Algorithms.” Paper to be presented at the International Workshop on Approximation Algorithms for Combinatorial Optimization Problems (APPROX), Paris, September 2016. Ghosh, Mrinalkanti, and Madhur Tulsiani. “From Weak to Strong LP Gaps for all CSPs.” Preprint, submitted August 1, 2016. arXiv:1608.00497v1. TALKS “The Constraint Satisfaction Problem.” Talk given at Dagstuhl workshop, July 2016. “A Characterization of Strong Approximation Resistance.” Invited talk given at International Symposium on Mathematical Programming (ISMP), August 2016. “Fourier Analysis.” Invited tutorial talk given at FSTTCS workshop, December 2016. INVOLVEMENT Editorial board member: Theory of Computing, Algorithmica Conference Reviews: STOC, FOCS, ICALP, CCC, TOC CLASSES/SEMINARS TTIC 31150 - Mathematical Toolkit (CMSC 31150) Autumn 2015: This is a new version of the 2013 course, which is now a list A course required for all TTIC students. The course focuses on various tools from linear algebra and probability required for research in computer science. Theory Reading Group.

20

Computational Biology Computational biology studies biological systems (e.g., cell, protein, DNA and RNA) through mathematical modeling and optimization. Machine learning methods (e.g., probabilistic graphical model and deep learning) and optimization techniques (e.g., linear programming and convex optimization) have significant applications in this field. Algorithm design and complexity analysis also play an important role, especially when we want to know if there is an efficient algorithm that can find an exact or approximate solution to a specific biological problem. Below is a list of the work done at TTIC this year in the area of Computational Biology.

Stefan Canzar Research Assistant Professor www.ttic.edu/canzar PUBLISHED/SUBMITTED PAPERS Canzar, Stefan, and Steven L. Salzberg. “Short Read Mapping: An Algorithmic Tour.” Proceedings of the IEEE PP, no. 99 (September 2015): 1-23. doi:10.1109/JPROC.2015.2455551. Canzar, Stefan, and Liliana Florea. “Computational Methods for Transcript Assembly from RNA-seq Reads.” In Computational Methods for Next Generation Sequencing Data Analysis 2016, ed. Ion Mandoiu and Alexander Zelikovsky, 199-216. Hoboken, NJ: WileyInterscience of John Wiley & Sons, Inc., 2016. Canzar, Stefan, Sandro Andreotti, David Weese, Knut Reinert, and Gunnar W. Klau. “CIDANE: Comprehensive Isoform Discovery and Abundance Estimation.” Genome Biology 17, no. 1 (January 2016): 1-18. doi:10.1186/s13059-015-0865-0. Canzar, Stefan, Khaled Elbassioni, Mitchell Jones, and Julian Mestre. “Resolving Conflicting Predictions from Multi-Mapping Reads.” Journal of Computational Biology 23, no. 3 (March 2016): 203-17. doi:10.1089/cmb.2015.0164. TALKS "Shedding Light on Invisible Transcripts by Algorithm Engineering." Invited talk given at Department of Informatics, Technical University of Munich, March 2016. “Computational Methods for Transcript Discovery and Quantification from RNA-seq Reads.” Talk given at Clinical and Translational Sciences Institute, Northwestern University, Chicago, September 2016. INVOLVEMENT PC member: RECOMB-Seq (Massive Parallel Sequencing) and RECOMB-CCB (Computational Cancer Biology) 2016, ECCB 2016, Bioinformatics 2016 Reviews: RECOMB 2016, ECCB 2016, and Bioinformatics 2016

Qixing Huang Research Assistant Professor www.ttic.edu/huang PUBLISHED/SUBMITTED PAPERS Hashemifar, Somaye, Qixing Huang, and Jinbo Xu. “Joint Alignment of Multiple Protein-Protein Interaction Networks via Convex Optimization.” Paper presented at the Annual International

21

Conference on Research in Computational Molecular Biology (RECOMB), Santa Monica, April 2016. doi:10.1089/cmb.2016.0025. Wei, Lingyu, Qixing Huang, Duygu Ceylan, Etienne Vouga, and Hao Li. “Dense Human Body Correspondences Using Convolutional Networks.” Paper presented at IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, June 2016. arXiv:1511.05904v2. Zhou, Tinghui, Philipp Krähenbühl, Mathieu Aubry, Qixing Huang, and Alexei A. Efros. “Learning Dense Correspondence via 3D-guided Cycle Consistency.” Paper presented at IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, June 2016. arXiv:1604.05383v1. Chen, Chen, Cewu Lu, Qixing Huang, Dimitrios Gunopulos, Leonidas Guibas, and Qiang Yang. “City-Scale Map Creation and Updating using GPS Collections.” Paper to be presented at the ACM SIGKDD Conference on Knowledge Discovery and Data Mining, San Francisco, August 2016. doi:10.1145/2939672.2939833. Wang, Ruizhe, Lingyu Wei, Etienne Vouga, Qixing Huang, Duygu Ceylan, Gerard Medioni, and Hao Li. “Capturing Dynamic Textured Surfaces of Moving Targets.” Paper to be presented at the European Conference on Computer Vision (ECCV), Amsterdam, October 2016. arXiv:1604.02801v1. Li, Yi, Vladimir Kim, Duygu Ceylan, I-Chao Shen, Mengyan Li, Hao Su, Cewu Lu, Qixing Huang, Alla Sheffer, and Leonidas Guibas. “A Scalable Active Framework for Region Annotation in 3D Shape Collections.” Paper to be presented at the ACM SIGGRAPH Conference and Exhibition on Computer Graphics and Interactive Techniques, Macao, December 2016. Shen, Yanyao, Qixing Huang, Nathan Srebro, and Sujay Sanghavi. “Normalized Spectral Map Synchronization.” Paper to be presented at the Annual Conference on Neural Information Processing Systems (NIPS), Barcelona, December 2016. Wang, Tuanfeng Y., Hao Su, Jinglei Huang, Qixing Huang, Leonidas Guibas, and Niloy Mitra. “Unsupervised Texture Transfer from Images to Model Collections.” Paper to be presented at the ACM SIGGRAPH Conference and Exhibition on Computer Graphics and Interactive Techniques, Macao, December 2016. Zhao, Haisen, Fanglin Gu, Qi-Xing Huang, Jorge Garcia, Yong Chen, Changhe Tu, Bedrich Benes, Hao Zhang, Daniel Cohen-Or, and Baoquan Chen. “Connected Fermat Spirals for Layered Fabrication.” Paper to be presented at the ACM SIGGRAPH Conference and Exhibition on Computer Graphics and Interactive Techniques, Macao, December 2016. doi:10.1145/2897824.2925958. TALKS “Visual Computing Using Big 3D Data.” Talk given at Computer Science Departmental Seminar, Purdue University, March 2016. “Visual Computing Using Big 3D Data.” Talk given at University of Texas at Austin, March 2016. “Visual Computing Using Big 3D Data.” Talk given at University of California at Riverside, March 2016. “Visual Computing Using Big 3D Data.” Talk given at Washington University at Saint Louis, March 2016. “Visual Computing Using Big 3D Data.” Talk given at University of Southern California, April 2016. “Dense Correspondences in the Era of Deep Learning.” Talk given as part of Research at TTIC, May 2016. INVOLVEMENT Program Committee Members: Symposium on Geometry Processing 2016, Shape Modeling International 2016 Conference Reviews: SIGGRAPH, SIGGRAPH Asia, CVPR, ECCV, PAMI, IJCV RESEARCH FUNDING AWARDS PI: GIFT Award. Intel Labs, October 2015- June 2016: $30,000.

22

NSF Award, Co-PI: AF: III: small: “Convex Optimization for Protein-Protein Interaction Network Alignment.” July 2016-June 2019. CLASSES/SEMINARS Two lectures in the online machine learning course for TTI (Japan) students.

Aly Khan Research Assistant Professor www.ttic.edu/khan PUBLISHED/SUBMITTED PAPERS Cho, Sunglim, Cheng-Jang Wu, Tomoharu Yasuda, Leilani O. Cruz, Aly A. Khan, Ling-Li Lin, Duc T. Nguyen, Marina Miller, Hyang-Mi Lee, et al. “miR-23 27 24 Clusters Control Effector T Cell Differentiation and Function.” Journal of Experimental Medicine 213, no. 2 (February 2016). doi:10.1083/jcb.2124OIA22. Canzar, Stefan, Karlynn E. Neu, Qingming Tang, Patrick C. Wilson, and Aly A. Khan. “BASIC: BCR Assembly from Single Cells.” Bioinformatics (forthcoming). TALKS “Computational Immunology: New Computational Approaches to Understand Immune Function.” Talk given at University of California, San Diego, August 2015. “Computational Immunology: New Computational Approaches to Understand Immune Function.” Talk given at Northwestern University, Evanston, May 2016. INVOLVEMENT Journal Reviews, Trends in Immunology, 2015. RESEARCH FUNDING AWARDS Deep Learning Hardware Grant, NVIDIA, October 2015. CLASSES/SEMINARS TTIC Reading Group participation: Bioinformatics and Computational Biology MISCELLANEOUS Thesis committee membership: Karlynn Neu, PhD Committee, University of Chicago Akinola Olumide Emmanuel, PhD Committee, University of Chicago Yuta Asano, PhD Committee, University of Chicago

Hammad Naveed Research Assistant Professor www.ttic.edu/naveed PUBLISHED/SUBMITTED PAPERS Cui, Xuefeng, Hammad Naveed, and Xin Gao. “Pairwise Structure Alignment Specifically Tuned for Surface Pockets and Interaction Interfaces.” Paper presented at ACM Conference on Bioinformatics, Computational Biology and Health Informatics, Atlanta, September 2015. doi:10.1145/2808719.2811431. Naveed, Hammad, Umar Hameed, Deborrah Harrus, William Bourguet, Stefan Arold, and Xin Gao. “An Integrated Structure- and System-Based Framework to Identify New Targets of Metabolites and Known Drugs.” Bioinformatics 31, no. 24 (September 2015): 3922-3929. doi:10.1093/bioinformatics/btv477.

23

Lin, Meishan, Dennis Gessmann, Hammad Naveed, and Jie Liang. “Outer Membrane Protein Folding and Topology from a Computational Transfer Free Energy Scale.” J American Chemical Society 138, no. 8 (February 2016): 2592-2601. doi:10.1021/jacs.5b10307. Liang, Jie, Youfang Cao, Gamze Gursoy, Hammad Naveed, Anna Terebus, and Jieling Zhao. “Multiscale Modeling of Cellular Epigenetic States: Stochasticity in Molecular Networks, Chromatin Folding in Cell Nuclei, and Tissue Pattern Formation of Cells.” Critical Reviews in Biomedical Engineering, forthcoming. doi:10.1615/CritRevBiomedEng.2016016559. TALKS “Inter-strand Contact Prediction for β-barrel Membrane Proteins.” Talk given at the International Conference on Protein and RNA Structure Prediction, Punta Cana, Dominican Republic, December 2015. “A Computational Framework for Predicting Novel Targets for Small-Molecules (Drugs).” Talk given at Computational Bioscience Research Center, KAUST, Saudi Arabia, January 2016. “An Integrated Structure and System-Based Framework to Identify Protein-Small Molecule Interactions.” Talk given at the IEEE International Conference on Biomedical and Health Informatics, Las Vegas, February 2016. INVOLVEMENT Preliminary exam committee, Jieling Zhao, PhD Candidate (Advisor: Jie Liang), Department of Bioengineering, UIC 2015 Co-organized and co-chaired a minisymposia titled “Computational precision medicine: prediction of cancer variants and druggable sites in proteins” at the BHI2016 IEEE International Conference on Biomedical and Health Informatics, Las Vegas, February 2016 Reviewer: Cogent Biology, IEEE/ACM Transactions on Computational Biology and Bioinformatics, and Annual EMBS Conference CLASSES/SEMINARS University of Illinois Chicago: BIOE 483: Molecular Modeling in Bioinformatics: Class for graduate students at the Bioinformatics Program of the Bioengineering Department during the Spring 2016 semester.

Jinbo Xu Associate Professor www.ttic.edu/xu PUBLISHED/SUBMITTED PAPERS Tang, Qingming, Chao Yang, Jian Peng, and Jinbo Xu. “Exact Hybrid Covariance Thresholding for Joint Graphical Lasso.” Paper presented at the Joint European Conference on Machine Learning and Knowledge Discovery in Databases (JCML), Porto, PT, September 2015. arXiv:1503.02128 Tang, Qingming, Sheng Wang, Jian Peng, Jianzhu Ma, and Jinbo Xu. “Bermuda: De Novo Assembly of Transcripts with New Insights for Handling Uneven Coverage.” Paper presented at the Association for Computing Machinery (ACM) Conference on Bioinformatics, Computational Biology and Health Informatics, Atlanta, GA, September 2015. arXiv:1506.05538 Sun, Siqi, Jianzhu Ma, Sheng Wang, and Jinbo Xu. “Predicting Diverse M-Best Protein Contact Maps.” Paper presented at the IEEE International Conference on Bioinformatics and Biomedicine (BIBM), Washington, DC, November 2015. doi:10.1109/BIBM.2015.7359865. Hashemifar, Somaye, Behnam Neyshabur, and Jinbo Xu. “Joint Inference of Tissue-Specific Networks with a Scale Free Topology.” Paper presented at the IEEE International Conference on Bioinformatics and Biomedicine (BIBM), Washington, DC, November 2015.

24

doi:10.1109/BIBM.2015.7359696. Sulakhe, Dinanath, Bingqing Xie, Andrew Taylor, Mark D’Souza, Sandhya Balasubramanian, Somaye Hashemifar, Steven White, Utpal J. Dave, Gady Agam, Jinbo Xu et al. “Lynx: A Knowledge Base and an Analytical Workbench for Integrative Medicine.” Nucleic Acids Research 44, no. D1 (January 2016): D882-D887. doi:10.1093/nar/gkv1257. Wang, Sheng, Jian Peng, Jianzhu Ma, and Jinbo Xu. “Protein Secondary Structure Prediction Using Deep Convolutional Neural Fields.” Scientific Reports 6, no. 18962 (January 2016). doi:10.1038/srep18962. Wang, Sheng, Wei Li, Shiwang Liu, and Jinbo Xu. “RaptorX-Property: A Web Server for Protein Structure Property Prediction.” Nucleic Acids Research (April 2016). doi:10.1093/nar/ gkw306. Hashemifar, Somaye, Qixing Huang, and Jinbo Xu. “Joint Alignment of Multiple Protein-Protein Interaction Networks via Convex Optimization.” Paper presented at the Annual International Conference on Research in Computational Molecular Biology (RECOMB), Santa Monica, April 2016. doi:10.1089/cmb.2016.0025. Wang, Sheng, Wei Li, Renyu Zhang, Shiwang Liu, and Jinbo Xu. “CoinFold: A Web Server for Protein Contact Prediction and Contact-Assisted Protein Folding.” Nucleic Acids Research (July 2016). doi:10.1093/nar/gkw307. Tang, Qingming, Lifu Tu, Weiran Wang, and Jinbo Xu. “Network Inference by Learned Node-Specific Degree Prior.” Preprint, submitted February 7, 2016. arXiv:1602.02386. TALKS “Probabilistic Graphical Models of Multiple Protein Sequence Alignment.” Talk given at University of Texas at Austin, November 2015. “Probabilistic Graphical Models of Multiple Protein Sequence Alignment.” Talk given at Colloquium, Tsinghua University, Beijing, December 2015. INVOLVEMENT Associate Editor, IEEE/ACM Trans. Bioinformatics and Computational Biology Program Committee Member: ACM BCB 2015, ISMB 2016, RECOMB 2016, and others Journal Reviews: PNAS, IEEE/ACM TCBB, PLoS Computational Biology, Bioinformatics, BMC Bioinformatics, Journal of Computational Biology, Proteins and Proteome HONORS/AWARDS Best Poster award, IJCAI BOOM Workshop, 2016 RESEARCH FUNDING AWARDS NIH/NIGMS 1R01GM089753-06A1. New Computational Methods for Data-Driven Protein Structure Prediction, September 2015-September 2019: $300,000/year. NSF DBI. ABI Development: Developing RaptorX Web Portal for Protein Structure and Functional Study, July 2016-June 2019: $180,000/year. NSF CCF. AF:III: small: Convex optimization for protein-protein interaction network alignment, July 2016 to June 2019: $100,000/year. NSF CAREER award. Exact and approximate algorithm for the 3D modeling of protein-protein interactions, 2012-2017: $100,000/year. CLASSES/SEMINARS TTIC 31050 - Introduction to Bioinformatics and Computational Biology: This course will focus on the application of mathematical models and computer algorithms to studying molecular biology. Bioinformatics reading group.

25

Computer Vision and Computational Photography Computer vision involves getting computers to extract useful information from pictures and videos. It has applications in robotics, surveillance, autonomous vehicles, and automobile collision avoidance. Historically, this is a central research area of computer science. Below is a list of the work done at TTIC this year in the area of Computer Vision and Computational Photography.

Ayan Chakrabarti Research Assistant Professor www.ttic.edu/chakrabarti PUBLISHED/SUBMITTED PAPERS “Color Constancy by Learning to Predict Chromaticity from Luminance.” Paper presented at the Neural Information Processing Systems Conference (NIPS), Montreal, December 2015. TALKS “Architectures for Learning in Low-Level Vision Applications.” Talk given as part of the Research at TTIC series, Chicago, January 2016. “A Neural Approach to Blind Motion Deblurring.” Talk given at the Midwest Vision Workshop, Chicago, April 2016. INVOLVEMENT Program Committee Member and Local Arrangements Chair, IEEE International Conference on Computational Photography (ICCP), 2016 Conference Reviews: NIPS 2015, CVPR 2016, ECCV 2016, ACCV 2016, SIGGRAPH Asia 2016, NIPS 2016 Workshop Reviews: Inverse Rendering (with ICCV 2015), Computational Cameras and Displays (with CVPR 2016) Journal Reviews: IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), IEEE Transactions on Computational Imaging (TCI), IEEE Transactions on Image Processing (TIP), SPIE Journal on Electronic Imaging (JEI), Computer Vision and Image Understanding (CVIU) RESEARCH FUNDING AWARDS Unrestricted Gift from Adobe Systems, August 2015: $5,000. Structured Inference for Low-Level Vision from NSF, June 2016-2019: $194,612. CLASSES/SEMINARS Vision Reading Group Deep Learning Reading Group

26

Michael Maire Research Assistant Professor www.ttic.edu/maire PUBLISHED/SUBMITTED PAPERS Narihira, Takuya, Michael Maire, and Stella X. Yu. “Direct Intrinsics: Learning Albedo-Shading Decomposition by Convolutional Regression.” Poster presented at the International Conference on Computer Vision (ICCV), Santiago, Chile, December 2015. doi:10.1109/ICCV.2015.342. Maire, Michael, Takuya Narihira, and Stella X. Yu. “Affinity CNN: Learning Pixel-Centric Pairwise Relations for Figure/Ground Embedding.” Paper presented at the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, June 2016. arXiv:1512.02767v2. INVOLVEMENT Reviews: NIPS 2015, IEEE TPAMI 2015, IEEE TNNLS 2015, IEEE TIP 2015, ECCV 2016. Area Chair, CVPR 2016

Greg Shakhnarovich Associate Professor www.ttic.edu/gregory PUBLISHED/SUBMITTED PAPERS Larsson, Gustav, Michael Maire, and Gregory Shakhnarovich. “Learning Representations for Automatic Colorization.” Paper to be presented at the European Conference on Computer Vision (ECCV), Amsterdam, October 2016. Chakrabarti, Ayan, Jingyu Shao, and Gregory Shakhnarovich. “Depth from a Single Image by Harmonizing Overcomplete Local Network Predictions.” Preprint, submitted May 23, 2016. arXiv:1605.07081. Larsson, Gustav, Michael Maire, and Gregory Shakhnarovich. “FractalNet: Ultra-Deep Neural Networks without Residuals.” Preprint, submitted May 24, 2016. arXiv:1605.07648. TALKS “Zoom-out Features for Image Understanding.” Talk given at University of Toronto: University of California at Berkeley: and University of Massachusetts Amherst, October 2016. “Rich Representations for Parsing Visual Scenes.” Talk given as part of VASC Seminar, Carnegie Mellon University, Pittsburgh, November 2016. “Rich Representations for Parsing Visual Scenes.” Talk given as part of AIIS Seminar, University of Illinois at Urbana-Champaign, December 2016. INVOLVEMENT Area Chair, CVPR 2016. Associate Editor, Computer Vision and Image Understanding Journal Organizer, Midwest Vision Workshop - April 2016 Conference Reviews: ICML, NIPS, ECCV RESEARCH FUNDING AWARDS Adobe Research Faculty Award, Spring 2016: $7,500. Spare5 Data Annotation Gift, Summer 2016: $25,000. CLASSES/SEMINARS TTIC 31020 - Introduction to Statistical Machine Learning: Graduate level introduction to principles and practice of machine learning. Vision Reading group (faculty coordinator): Weekly reading group focused on current literature in the area of computer vision.

27

Deep Learning Discussion group (organizer): Weekly discussion group focused on topics related to deep learning, across many application domains relevant to TTIC (speech, vision, NLP). MISCELLANEOUS Director of Admissions

Machine Learning Machine Learning generally refers to an engineering or design paradigm where systems are built based on automatic training from examples, rather than detailed expert knowledge, much in the same way humans learn how to perform tasks and interact with the world. Most of modern Machine Learning is statistical in nature, and builds on statistical and probabilistic tools, as well as on algorithmic and computational developments. Especially in recent years, as training data is becoming plentiful, and massive computational and storage resources needed for handling the data are also becoming available, Machine Learning is playing a key role in many application areas. These include both classic Artificial Intelligence problems, such as computer vision, robotics, machine translation, question answering and dialogue systems. There are also a variety of “non-human” problems such as information retrieval, search, bioinformatics and stock market prediction to be considered. Below is a list of the work done at TTIC this year in the area of Machine Learning.

Srinadh Bhojanapalli Research Assistant Professor www.ttic.edu/bhojanapalli PUBLISHED/SUBMITTED PAPERS Bhojanapalli, Srinadh, Anastasios Kyrillidis, and Sujay Sanghavi. “Dropping Convexity for Faster Semidefinite Optimization.” Paper presented at the OPT-ML Workshop at the Advances in Neural Information Processing Systems (NIPS), Montreal, Canada, December 2015. Chen, Yudong, Srinadh Bhojanapalli, Sujay Sanghavi, and Rachel Ward. “Completing Any Lowrank Matrix, Provably.” Journal of Machine Learning Research (JMLR) 16 (December 2015). arXiv:1306.2979v4. Bhojanapalli, Srinadh, Anastasios Kyrillidis, and Sujay Sanghavi. “Dropping Convexity for Faster Semidefinite Optimization.” Paper presented at Conference on Learning Theory (COLT), New York City, June 2016. arXiv:1509.03917v3. TALKS “Dropping Convexity for Faster Semidefinite Optimization.” Talk given at Signals, Inference, and Networks (SINE) Seminar, University of Illinois at Urbana-Champaign, October 2015. INVOLVEMENT Conference Reviews: NIPS 2015, ICML 2016, COLT 2016, AISTATS 2016 Journal Reviews: JMLR, SIAM Review, IEEE Journal of Selected Topics in Signal Processing. Organized "Advances in Non-convex Analysis and Optimization" workshop at ICML 2016

28

Dan Garber Research Assistant Professor www.ttic.edu/garber PUBLISHED/SUBMITTED PAPERS Garber, Dan, Elad Hazan, Chi Jin, Sham M. Kakade, Cameron Musco, Praneeth Netrapalli, and Aaron Sidford. “Faster Eigenvector Computation via Shift-and-Invert Preconditioning.” Paper presented at the International Conference on Machine Learning (ICML), New York, June 2016. TALKS “Projection-free Optimization and Learning.” Talk given at Operations Research Seminar, Technion-Israel Institute of Technology, Haifa, Israel, April 2016. “Fast and Simple PCA via Convex Optimization.” Talk given at Optimization and Statistical Learning Seminar, Northwestern University, Evanston, April 2016. INVOLVEMENT Journal Reviews: Journal of Machine Learning Research (JMLR); Journal of Mathematical Programming; Neural Information Processing Systems (NIPS) 2016

Mehrdad Mahdavi Research Assistant Professor www.ttic.edu/mahdavi PUBLISHED/SUBMITTED PAPERS Meshi, Ofer, Mehrdad Mahdavi, and Alex Schwing. “Smooth and Strong: MAP Inference with Linear Convergence.” Poster presented at Neural Information Processing Systems (NIPS), Montreal, Canada, December 2015. Meshi, Ofer, Mehrdad Mahdavi, Adrian Weller, and David Sontag. “Train and Test Tightness of LP Relaxations in Structured Prediction.” Paper presented at the International Conference on Machine Learning (ICML), New York, June 2016. doi:10.17863/CAM.242. TALKS “Lower and Upper Bounds on the Generalization of Stochastic Exponentially Concave Optimization.” Talk given at Data Science Institute, Columbia University, New York, August 2015. “Randomized Algorithms for Large-Scale Learning: From Optimization to Recovery.” Talk given at Center for Statistics and Machine Learning, Princeton University, NJ, November 2015. “Stochastic Optimization for High-Dimensional Large-Scale Learning.” Talk given at Wisconsin Institute for Discovery, University of Wisconsin-Madison, March 2016.

David McAllester Chief Academic Officer, Professor www.ttic.edu/mcallester TALKS “An Overview of Morphoid Type Theory.” Talk given at Purdue University, West Lafayette, IN, March 2016.

29

“An Overview of Morphoid Type Theory.” Talk given at Indiana University, Bloomington, IN, March 2016. “An Overview of Morphoid Type Theory.” Talk given at Brown University, Providence, RI, April 2016. “Strong AI, Prospects and Control.” Talk given at Brown University, Providence, RI, April 2016. “An Overview of Morphoid Type Theory.” Talk given at Cornell University, Ithaca, NY, May 2016. CLASSES/SEMINARS TTIC 31040 Introduction to Computer Vision (CMSC 35040): Introduction to deep learning for computer vision. Although deep learning based computer vision systems are evolving rapidly, this course attempts to teach material that will remain relevant and useful as the field changes. The course begins with general deep learning methods relevant to many applications and gradually focuses to a greater extent on computer vision. The course will emphasize theoretical and intuitive understanding to the extent possible. Reading group on machine comprehension (with Kevin Gimpel, Mohit Bansal, Hai Wang, Takeshi Onishi).

Ofer Meshi Research Assistant Professor www.ttic.edu/meshi PUBLISHED/SUBMITTED PAPERS Meshi, Ofer, Mehrdad Mahdavi, and Alex Schwing. “Smooth and Strong: MAP Inference with Linear Convergence.” Poster presented at Neural Information Processing Systems (NIPS), Montreal, Canada, December 2015. Meshi, Ofer, Mehrdad Mahdavi, and David Sontag. “On the Tightness of LP Relaxations for Structured Prediction.” Paper presented at NIPS Workshop on Optimization for Machine Learning, Montreal, Canada, December 2015. arXiv:1511.01419v2. Choi, Heejin, Ofer Meshi, and Nathan Srebro. “Fast and Scalable Structural SVM with Slack Rescaling.” Poster presented at the Artificial Intelligence and Statistics (AISTATS), Fort Lauderdale, May 2016. Meshi, Ofer, Mehrdad Mahdavi, Adrian Weller, and David Sontag. “Train and Test Tightness of LP Relaxations in Structured Prediction.” Paper presented at the International Conference on Machine Learning (ICML), New York, June 2016. doi:10.17863/CAM.242. Garber, Dan, and Ofer Meshi. “Linear-Memory and Decomposition-Invariant Linearly Convergent Conditional Gradient Algorithm for Structured Polytopes.” Preprint, submitted May 20, 2016. arXiv:1605.06492v1. TALKS “Scalable Machine Learning for High-Dimensional Structured Outputs.” Talk given at Tel Aviv University, Israel, January 2016. “Scalable Machine Learning for High-Dimensional Structured Outputs.” Talk given at Bar Ilan University, Ramat Gan, Israel, January 2016. “Scalable Machine Learning for High-Dimensional Structured Outputs.” Talk given at Technion, Haifa, Israel, January 2016. “Scalable Machine Learning for High-Dimensional Structured Outputs.” Talk given at University of Illinois Urbana-Champaign, IL, March 2016. “Scalable Machine Learning for High-Dimensional Structured Outputs.” Talk given at University of Maryland, College Park, MD, March 2016. INVOLVEMENT Conference reviews: AISTATS 2016, ICML 2016, UAI 2016. Journal reviews: Computer Vision and Image Understanding (CVIU), IEEE Transactions on Pattern Analysis and Machine Intelligence.

30

CLASSES/SEMINARS TTIC 31070 - Convex Optimization: Joint with Nathan Srebro: The course will cover techniques in unconstrained and constrained convex optimization and a practical introduction to convex duality. The course will focus on (1) formulating and understanding convex optimization problems and studying their properties; (2) understanding and using the dual; and (3) presenting and understanding optimization approaches, including interior point methods and first order methods for non-smooth problems. Examples will be mostly from data fitting, statistics and machine learning. TTIC 31180 - Probabilistic Graphical Models, Spring 2016 - two lectures.

Nathan Srebro Associate Professor www.ttic.edu/srebro PUBLISHED/SUBMITTED PAPERS Neyshabur, Behnam, Ryota Tomioka, and Nathan Srebro. “Norm-Based Capacity Control in Neural Networks.” Paper presented at the Conference on Learning Theory (COLT), Paris, July 2015. arXiv:1503.00036v2. Neyshabur, Behnam, Nathan Srebro. “On Symmetric and Asymmetric LSHs for Inner Product Search.” Paper presented at the the International Conference on Machine Learning (ICML), Lille, France, July 2015. arXiv:1410.5518v3. Wang, Jialei, Mladen Kolar, Nathan Srebro. “Distributed Multitask Learning.” Paper at the Annual Allerton Conference, University of Urbana-Champaign, September 2015. arXiv:1510.00633v1. Wang, Weiran, Raman Arora, Karen Livescu, and Nathan Srebro. “Stochastic Optimization for Deep CCA via Nonlinear Orthogonal Iterations.” Paper presented at the Annual Allerton Conference, University of Urbana-Champaign, September 2015. doi:10.1109/ ALLERTON.2015.7447071. Neyshabur, Behnam, Ruslan R. Salakhutdinov, Nathan Srebro. “Path-SGD: Path-Normalized Optimization in Deep Neural Networks.” Paper presented at the Advances in Neural Information Processing Systems (NIPS), Montreal, Canada, December 2015. arXiv:1506.02617v1. Needell, Deanna, Nathan Srebro, and Rachel Ward. “Stochastic Gradient Descent, Weighted Sampling, and the Randomized Kaczmarz Algorithm.” Mathematical Programming 155 (January 2016):549-573. doi:10.1007/s10107-015-0864-7. Choi, Heejin, Ofer Meshi, and Nathan Srebro. “Fast and Scalable Structural SVM with Slack Rescaling.” Paper presented at the International Conference on Artificial Intelligence and Statistics (AISTATS), Fort Lauderdale, May 2016. arXiv:1510.06002v2. Neyshabur, Behnam, Ryota Tomioka, Ruslan R. Salakhutdinov, Nathan Srebro. “DataDependent Path Normalization in Neural Networks.” Paper presented at the International Conference on Learning Representations (ICLR), San Juan, Puerto Rico, May 2016. arXiv:1511.06747v4. Wang, Jialei, Mladen Kolar, Nathan Srebro. “Distributed Multi-Task Learning with Shared Representation.” Paper presented at the International Conference on Artificial Intelligence and Statistics (AISTATS), Fort Lauderdale, May 2016. arXiv:1603.02185v1. Bijral, Avleen S., Anand D. Sarwate, and Nathan Srebro. “On Data Dependence in Distributed Stochastic Optimization.” IEEE Transactions on Automatic Control (forthcoming). arXiv:1603.04379v1. Takáč, Martin, Peter Richtárik, and Nathan Srebro. “Distributed Mini-Batch SDCA.” Preprint, submitted July 29, 2015. arXiv:1507.08322v1. Wang, Jialei, Hai Wang, and Nathan Srebro. “Reducing Runtime by Recycling Samples.” Preprint, submitted February 5, 2016. arXiv:1602.02136v1.

31

Choi, Heejin, Yutaka Sasaki, and Nathan Srebro. “Normalized Hierarchical SVM.” Preprint, submitted March 4, 2016. arXiv:1508.02479v2. Wang, Weiran, Jialei Wang, and Nathan Srebro. “Globally Convergent Stochastic Optimization for Canonical Correlation Analysis.” Preprint, submitted May 20, 2016. arXiv:1604.01870v3. Bhojanapalli, Srinadh, Behnam Neyshabur, and Nathan Srebro. “Global Optimality of Local Search for Low Rank Matrix Recovery.” Preprint, submitted May 23, 2016. arXiv:1605.07221v2. Neyshabur, Behnam, Yuhuai Wu, Ruslan Salakhutdinov, and Nathan Srebro. “Path-Normalized Optimization of Recurrent Neural Networks with ReLU Activations.” Preprint, submitted May 23, 2016. arXiv:1605.07154v1. Wang, Jialei, Mladen Kolar, Nathan Srebro, and Tong Zhang. “Efficient Distributed Learning with Sparsity.” Preprint, submitted May 25, 2016. arXiv:1605.07991v1. Woodworth, Blake, and Nathan Srebro. “Tight Complexity Bounds for Optimizing Composite Objectives.” Preprint, submitted May 25, 2016. arXiv:1605.08003v1. TALKS “Regularization, Optimization and Generalization in MultiLayer Networks.” Talk given as part of the Dagstuhl Workshop on Mathematical and Computational Foundations of Learning Theory, Germany, September 2015. “Optimization, Regularization and Generalization in Multilayer Networks.” Talk given at Hebrew University, Jerusalem; and Weizmann Institute, Rehovot, Israel, November 2015. “Learning, Stability and Strong Convexity.” Invited talk at NIPS Workshop on Adaptive Data Analysis, Montreal, December 2015. “Acceleration in Distributed Stochastic Optimization.” Talk given at the Optimization Without Borders Workshop, Les Houches, France, February 2016. “Distributed Multi-Task Learning.” Talk in Invited Session, ITA Workshop, La Jolla, February 2016. “Geometry, Optimization and Regularization in Deep Learning.” Invited talk at ICRIRI, May 2016. “Supervised Learning without Discrimination.” Talk given at Hebrew University, Jerusalem, May 2016. “The Power of Asymmetry in Binary Hashing.” Talk given at TTIJ Joint CS Seminar, Nagoya, September 2016. INVOLVEMENT Two NSF Panels, 2015. Action Editor, Journal of Machine Learning Research (JMLR). Area Chair: NIPS 2015, ICML 2016. Senior Program Committee, COLT 2016. IJCAI Program Committee. ICLR Program Committee. RESEARCH FUNDING AWARDS NSF BIG-DATA, four years: $1,500,000. CLASSES/SEMINARS TTIC 31070 - Convex Optimization: The course will cover techniques in unconstrained and constrained convex optimization and a practical introduction to convex duality. The course will focus on (1) formulating and understanding convex optimization problems and studying their properties; (2) understanding and using the dual; and (3) presenting and understanding optimization approaches, including interior point methods and first order methods for nonsmooth problems. Examples will be mostly from data fitting, statistics and machine learning. Machine Learning and Optimization Reading Group.

32

Robotics Robotics can generally be defined as a field concerned with the development and realization of intelligent, physical agents that are able to perceive, plan, and act intentionally in an uncertain world. Robotics is a broad field that includes mechanical design, planning and control, perception, estimation, and human-robot interaction, among others. At TTIC, robotics research currently focuses on developing advanced perception algorithms that endow robots with a rich awareness of, and the ability to act deliberately within, their surroundings. Researchers are particularly interested in algorithms that take multi-modal observations of a robot’s surround as input, notably image streams and natural language speech, and infer rich properties of the people, places, objects, and actions that comprise a robot’s environment. Integral to these technologies is their reliance on techniques from machine learning in developing probabilistic and statistical methods that are able to overcome the challenge of mitigating the uncertainty inherent in performing tasks effectively in real-world environments. These tasks include assistive technology for people living with physical and cognitive impairments, healthcare, logistics, manufacturing, and exploration. Below is a list of the work done at TTIC this year in the area of Robotics.

Matthew Walter Assistant Professor www.ttic.edu/walter PUBLISHED/SUBMITTED PAPERS Chung, Istvan, Oron Propp, Matthew R. Walter, and Thomas M. Howard. “On the Performance of Hierarchical Distributed Correspondence Graphs for Efficient Symbol Grounding of Robot Instructions.” Paper presented at the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Hamburg, 2015. doi:10.1109/IROS.2015.7354117. Hemachandra, Sachithra, Matthew R. Walter, and Seth Teller. “Information Theoretic Question Asking to Improve Spatial Semantic Representations.” Paper presented at the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Hamburg, 2015. doi:10.1109/IROS.2015.7354097. Chu, Hang, Hongyuan Mei, Mohit Bansal, and Matthew R. Walter. “Accurate Vision-Based Localization by Transferring Between Ground and Satellite Images.” Paper presented at the Advances in Neural Information Processing Systems (NIPS) Workshop on Transfer and Multi-Task Learning, Montreal, Canada, December 2015. arXiv:1510.09171v1. Fakoor, Rasool, Mohit Bansal, and Matthew R. Walter. “Deep Attribute-based Zero-shot Learning with Layer-specific Regularizers.” Paper presented at the Advances in Neural Information Processing Systems (NIPS) Workshop on Transfer and Multi-Task Learning, Montreal, Canada, December 2015. Mei, Hongyuan, Mohit Bansal, and Matthew R. Walter. “Listen, Attend, and Walk: Neural Mapping of Navigational Instructions to Action Sequences.” Paper presented at the Advances in Neural Information Processing Systems (NIPS) Workshop on Multimodal and Machine Learning, Montreal, Canada, December 2015. Mei, Hongyuan, Mohit Bansal, and Matthew R. Walter. “Listen, Attend, and Walk: Neural Mapping of Navigational Instructions to Action Sequences.” Paper presented at the National

33

Conference on Artificial Intelligence (AAAI), Phoenix, AZ, February 2016. arXiv:1506.04089v4. Barber, Daniel J., Thomas M. Howard, and Matthew R. Walter. “A Multimodal Interface for Real-Time Soldier-Robot Teaming.” Proceedings of SPIE, Unmanned Systems Technology, May 13, 2016. doi:10.1117/12.2224401. Daniele, Andrea F., Mohit Bansal, and Matthew R. Walter. "Natural Language Generation in the Context of Providing Indoor Route Instructions." Paper presented as part of Robotics: Science and Systems Workshop on Model Learning for Human-Robot Communication, University of Michigan, June 2016. Mei, Hongyuan, Mohit Bansal, and Matthew R. Walter. “What to talk about and how? Selective Generation using LSTMs with Coarse-to-Fine Alignment.” Paper presented at the Conference of the North American Chapter of the Association for Computational Linguistics of Human Language Technologies (NAACL HLT), San Diego, June 2016. arXiv:1509.00838v2. Oh, Jean H., Thomas M. Howard, Matthew R. Walter, and Daniel J. Barber. “Integrated Intelligence for Human-Robot Teams.” Paper to be presented at the International Symposium on Experimental Robotics (ISER), Tokyo, October 2016. TALKS “Real-Time Analytics Onboard Self-Driving Cars: Perception-Driven Autonomous Vehicles.” Talk given at Booth School of Business, University of Chicago, August 2015. “Smart Cars: Perception-Driven Autonomous Vehicles,” Talk given at the Technology Leaders Association (TLA) Managers Forum, September 2015. “Smart Cars: Perception-Driven Autonomous Vehicles.” Talk given at Northwestern University, February 2016. “Smart Cars: Perception-Driven Autonomous Vehicles,” Talk given at Chicago City Data User Group, April 2016. “Information-Theoretic Question Asking to Improve Spatial-Semantic Representations.” Talk given at the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), October 2015. “Following Natural Language Instructions in Unknown Environments.” Talk given at Department of Electrical Engineering and Computer Science, University of Michigan, February 2016. “Following Natural Language Instructions in Unknown Environments.” Talk given at University of Wisconsin, Madison, March 2016. “How Do Robots Think?” Talk given at Girls Who Code Summer Immersion Program, July 2016. INVOLVEMENT Senior Editor, Spatial Reasoning and Interaction for Real-World Robotics Special issue, RSJ Advanced Robotics Organizer: 2016 Robotics: Science and Systems Workshop on Model Learning for HumanRobot Communication, first Midwest Robotics Workshop Associate Editor, IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) Panelist, NSF National Robotics Initiative Czar: TTIC 2015/2016, 2016/2017 Young Researcher Seminar Series, TTIC faculty recruiting. Program Committee, 2016 Conference on Empirical Methods in Natural Language Processing (EMNLP) Steering Committee, Northeast Robotics Colloquium Reviewer: IEEE Transactions on Robotics, International Journal of Robotics Research, Autonomous Robots, Journal of Translational Engineering in Health and Medicine, IEEE International Conference on Robotics and Automation (ICRA), IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), ACM/IEEE International Conference on Human-Robot Interaction (HRI) HONORS/AWARDS Best Paper Award, 2015 NIPS Workshop on Multimodal Machine Learning

34

RESEARCH FUNDING AWARDS Office of Naval Research (ONR), “Object Detection and Reacquisition from Visual and Lingual Signals via Zero-Shot Learning and Spatial- Semantic Mapping:” $160,000 (Pending). Army Research Laboratory (ARL), Robotics Collaborative Task Alliance (RCTA), “Learning Task Ordering from Natural Language Directions:” $32,000. Army Research Laboratory (ARL), Robotics Collaborative Task Alliance (RCTA), “Learning to Handle Objects from Human Demonstrations:” $45,000. CLASSES/SEMINARS TTIC 31180 - Probabilistic Graphical Models: This graduate-level course will provide a strong foundation for learning and inference with probabilistic graphical models. The course will first introduce the underlying representational power of graphical models, including Bayesian and Markov networks, and dynamic Bayesian networks. Next, the course will investigate contemporary approaches to statistical inference, both exact and approximate. The course will then survey state-of-the-art methods for learning the structure and parameters of graphical models. Robotics Reading Group. Rapid Robotics: Autonomous Systems with Open Source Software, Massachusetts Institute of Technology: Thanks to open source libraries and inexpensive robot platforms, creating advanced robot capabilities has never been more accessible. This course is a hands-on introduction to applied robotics software programming. You will learn to use the popular ROS robotics framework, open source autonomy libraries, and a small ground robot equipped with an RGB-depth sensor to demonstrate behaviors such as person-following, patrolling, exploration, and map-making. Lectures accompanying the laboratory exercises will cover the basics of robotics and autonomy algorithm theory. Participants will work in teams of two on robot systems. MISCELLANEOUS Supervisor, Hongyuan Mei, M.S. Physical Sciences, University of Chicago, 2016 Supervisor, Bharat Chandar, M.S. Statistics, University of Chicago, 2016 Supervisor, Andrea F. Daniele, Sapienza, Universit`a di Roma, 2016 Supervisor, Zhongtian Dai, TTIC PhD Committee Member, Jianzhu Ma, TTIC Qualifying Exam Committee Member, Mohammadreza Mostajabi, TTIC Qualifying Exam Committee Member, Hai Wang, TTIC

Speech and Language Technologies This area is concerned with getting computers to analyze and extract information from spoken language, as well as to generate spoken audio. At TTIC, current speech research focuses mainly on the analysis side. For example, speech recognition is the problem of transcribing the words being spoken in an audio signal, such as that recorded from a microphone. Speech processing heavily relies on techniques from machine learning and statistics, as well as ideas from linguistics and speech science, and shares algorithms with computer vision and computational biology. This area has applications such as automated telephone information centers, dictation systems, machine translation, archiving and search of spoken documents, assistance for the visually or hearingimpaired, and other human-computer interface systems. Below is a list of the work done at TTIC this year in the area of Speech and Language Technologies.

35

Mohit Bansal Research Assistant Professor www.ttic.edu/bansal PUBLISHED/SUBMITTED PAPERS Mei, Hongyuan, Mohit Bansal, and Matthew R. Walter. “Listen, Attend, and Walk: Neural Mapping of Navigational Instructions to Action Sequences.” Paper presented at the National Conference on Artificial Intelligence (AAAI), Phoenix, AZ, February 2016. arXiv:1506.04089v4. Wieting, John, Mohit Bansal, Kevin Gimpel, and Karen Livescu. “Towards Universal Paraphrastic Sentence Embeddings.” Paper presented at the International Conference on Learning Representations (ICLR), San Juan, Puerto Rico, May 2016. arXiv:1511.08198. Chandrasekaran, Arjun, Ashwin Kalyan, Stanislaw Antol, Mohit Bansal, Dhruv Batra, C. Lawrence Zitnick, and Devi Parikh. “We Are Humor Beings: Understanding and Predicting Visual Humor.” Paper presented at IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, June 2016. arXiv:1512.04407v4. Mei, Hongyuan, Mohit Bansal, and Matthew R. Walter. “What to Talk About and How? Selective Generation using LSTMs with Coarse-to-Fine Alignment.” Paper presented at the Conference of the North American Chapter of the Association for Computational Linguistics of Human Language Technologies (NAACL HLT), San Diego, June 2016. arXiv:1509.00838v2. Melamud, Oren, David McClosky, Siddharth Patwardhan, and Mohit Bansal. “The Role of Context Types and Dimensionality in Learning Word Embeddings.” Paper presented at the Conference of the North American Chapter of the Association for Computational Linguistics (NAACL), San Diego, June 2016. rXiv:1601.00893v1. Madhyastha, Pranava S., Mohit Bansal, Kevin Gimpel, and Karen Livescu. “Mapping Unseen Words to Task-Trained Embedding Spaces.” Paper to be presented as part of Workshop on Representation Learning for NLP, Berlin, August 2016. arXiv:1510.02387v2. Miwa, Makoto, and Mohit Bansal. “End-to-End Relation Extraction using LSTMs on Sequences and Tree Structures.” Paper to be presented at the Association for Computational Linguistics (ACL), Berlin, August 2016. arXiv:1601.00770v2. TALKS “Neural Attention Models for Natural Language Grounding and Generation.” Talk given at IIT Kanpur, India, October 2015. “Neural Attention Models for Natural Language Grounding and Generation.” Talk given at IIT Danpur, India, October 2015. “Structured Learning of World Knowledge for Natural Language Semantics.” Talk given at University of California, Davis, April 2016. “Structured Learning of World Knowledge for Natural Language Semantics.” Talk given at Rutgers University, New Brunswick, NJ, April 2016. “Structured Learning of World Knowledge for Natural Language Semantics.” Talk given at University of Texas at Austin, March 2016. “Structured Learning of World Knowledge for Natural Language Semantics.” Talk given at University of North Carolina at Chapel Hill, March 2016. “Structured Learning of World Knowledge for Natural Language Semantics.” Talk given at Carnegie Mellon University, Pittsburgh, PA, March 2016. “Structured Learning of World Knowledge for Natural Language Semantics.” Talk given at University of California Irvine, March 2016. “Structured Learning of World Knowledge for Natural Language Semantics.” Talk given at MSR, March 2016. “Structured Learning of World Knowledge for Natural Language Semantics.” Talk given at Virginia Tech, Blacksburg, VA, February 2016. INVOLVEMENT Tutorial Co-chair, NAACL 2016

36

Area Co-chair, NAACL 2016 Journal Reviews: TACL, TPAMI Conference Reviews: EMNLP, NAACL, ACL, NIPS, ICLR, IJCAI Demo Co-chair, ACL 2017 HONORS/AWARDS NVidia Paper Awards, “Listen, Attend, and Walk: Neural Mapping of Navigational Instructions to Action Sequences.” NIPS 2015 Multimodal Machine Learning Workshop. RESEARCH FUNDING AWARDS NVidia Hardware Grant, 2015. Bloomberg Data Science Research Grant, 2016: $60,000.

Kevin Gimpel Assistant Professor www.ttic.edu/gimpel PUBLISHED/SUBMITTED PAPERS Wieting, John, Mohit Bansal, Kevin Gimpel, Karen Livescu, and Dan Roth. “From Paraphrase Database to Compositional Paraphrase Model and Back.” Transactions of the Association for Computational Linguistics 3 (2015). https://transacl.org/ojs/index.php/tacl/article/ view/571. Wang, Hai, Mohit Bansal, Kevin Gimpel, and David McAllester. “Machine Comprehension with Syntax, Frames, and Semantics.” Poster presented at the Annual Meeting of the Association for Computational Linguistics (ACL), Beijing, July 2015. He, Hua, Kevin Gimpel, and Jimmy Lin. “Multi-Perspective Sentence Similarity Modeling with Convolutional Neural Networks.” Poster presented at Conference on Empirical Methods in Natural Language Processing (EMNLP), Lisbon, September 2015. Tang, Hao, Weiran Wang, Kevin Gimpel, and Karen Livescu. “Discriminative Segmental Cascades for Feature-Rich Phone Recognition.” Poster presented at IEEE Automatic Speech Recognition and Understanding Workshop, Scottsdale, December 2015. Wieting, John, Mohit Bansal, Kevin Gimpel, and Karen Livescu. “Towards Universal Paraphrastic Sentence Embeddings.” Paper presented at the International Conference on Learning Representations (ICLR), San Juan, Puerto Rico, May 2016. arXiv:1511.08198. He, Hua, John Weiting, Kevin Gimpel, Jinfeng Rao, and Jimmy Lin. “UMD-TTIC-UW at SemEval -2016 Task 1: Attention-Based Multi-Perspective Convolutional Neural Networks for Textual Similarity Measurement.” Paper presented at the International Workshop on Semantic Evaluation (SemEval), San Diego, June 2016. doi:10.18653/v1/S16-1170. TALKS “Discriminative Training.” Invited talk given at Second Machine Translation Marathon in the Americas, Notre Dame, IN, May 2016. INVOLVEMENT Area Co-chair, NAACL 2016 Journal Reviews: Transactions of the Association for Computational Linguistics, Open Linguistics Conference Reviews: AAAI, ACL, CoNLL, EMNLP, ICLR, MT Summit, SemEval HONORS/AWARDS Nomination, Best Paper Award, IEEE Automatic Speech Recognition and Understanding Workshop, 2015.

37

Ranked fifth out of forty, Best of SemEval, 2016. RESEARCH FUNDING AWARDS Microsoft Azure Sponsorship Offer, May 2016-2017: $20,000. Argonne Leadership Computing Facility Allocation, May 2016-October 2016: 10,000 computehours. CLASSES/SEMINARS TTIC 31190: Natural Language Processing: Introductory course covering fundamental concepts, problems, and methods in natural language processing.

Karen Livescu Assistant Professor www.ttic.edu/livescu PUBLISHED/SUBMITTED PAPERS Wang, Weiran, Raman Arora, Karen Livescu, and Nathan Srebro. “Stochastic Optimization for Deep CCA via Nonlinear Orthogonal Iterations.” Paper presented at the Annual Allerton Conference, University of Urbana-Champaign, September 2015. doi:10.1109/ ALLERTON.2015.7447071. Tang, Hao, Weiran Wang, Kevin Gimpel, and Karen Livescu. “Discriminative Segmental Cascades for Feature-Rich Phone Recognition.” Poster presented at the IEEE Workshop on Automatic Speech Recognition and Understanding (ASRU), Scottsdale, AZ, December 2015. Kamper, Herman, Weiran Wang, and Karen Livescu. “Deep Convolutional Acoustic Word Embeddings Using Word-Pair Side Information.” Paper presented at the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Shanghai, China, March 2016. doi:10.1109/ICASSP.2016.7472619. Kim, Taehwan, Weiran Wang, Hao Tang, and Karen Livescu. “Signer-independent Fingerspelling Recognition with Deep Neural Network Adaptation.” Poster presented at the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Shanghai, China, March 2016. Livescu, Karen, Preethi Jyothi, and Eric Fosler-Lussier. “Articulatory Feature-based Pronunciation Modeling.” Computer Speech and Language 36 (March 2016): 212–232. doi:10.1016/j.csl.2015.07.003. Wang, Weiran, and Karen Livescu. “Large-Scale Approximate Kernel Canonical Correlation Analysis.” Paper and poster presented at the International Conference on Learning Representations (ICLR), San Juan, Puerto Rico, May 2016. arXiv:1511.04773v4. Wieting, John, Mohit Bansal, Kevin Gimpel, and Karen Livescu. “Towards Universal Paraphrastic Sentence Embeddings.” Paper presented at the International Conference on Learning Representations (ICLR), San Juan, Puerto Rico, May 2016. arXiv:1511.08198. Michaeli, Tomer, Weiran Wang, and Karen Livescu. “Nonparametric Canonical Correlation Analysis." Paper presented at the International Conference on Machine Learning (ICML), New York, June 2016. arXiv:1511.04839v4. TALKS “Segmental Models in the Neural Age.” Talk given at LTI Colloquium, Carnegie Mellon University, Pittsburgh, PA, December 2015. “Segmental Models in the Neural Age.” Talk given at DSP Seminar, Technion, Haifa, Israel, January 2016. “Segmental Models in the Neural Age.” Talk given at Northwestern CS Division Seminar, Northwestern University, Evanston, IL, May 2016.

38

“Multi-view Learning of Representations for Speech and Language.” Talk given at Statistics Department Colloquium, University of Chicago, IL, February 2016. “Multi-view Learning of Representations for Speech and Language.” Talk given at JHU Summer School on Human Language Technology, John Hopkins University, Baltimore, Maryland, June 2016. “Triphone State Tying via Deep Canonical Correlation Analysis.” Talk given at Midwest Speech and Language Days, Indiana University, Bloomington, May 2016. “Deep Convolutional Word Embeddings with Word-Pair Side Information.” Talk given at Midwest Speech and Language Days, Indiana University, Bloomington, May 2016. “Acoustic Word Embeddings and Neural Segmental Models.” Talk given at Columbia University Speech Group, New York, June 2016. “Acoustic Word Embeddings and Neural Segmental Models.” Talk given at JHU Summer School on Human Language Technology, John Hopkins University, Baltimore, Maryland, June 2016. INVOLVEMENT Technical Co-chair, ASRU 2015 Associate Editor, IEEE Transactions on Audio, Speech, and Language Processing. Area Chair, ICASSP 2016 Conference Reviews: Interspeech 2015, ASRU 2015, ICLR 2016 HONORS/AWARDS Best Paper Nominee, ASRU 2015. Best Student Paper of Speech and Language Processing, ICASSP 2016. RESEARCH FUNDING AWARDS Google Research Award: "Discriminative Neural Segmental Models for Sequence Prediction." $62,004. February 2016 for one year. PI: Karen Livescu. CLASSES/SEMINARS TTIC 31110: Speech Technologies: This course introduces techniques used in speech technologies, mainly focusing on speech recognition. Speech recognition is one of the oldest and most complex structured sequence prediction tasks receiving significant research and commercial attention, and therefore provides a good case study for many of the techniques that are used in other areas of artificial intelligence involving sequence modeling. It is also a good example of the effectiveness of combining statistics and learning with domain knowledge. The course includes practical homework exercises using Matlab and speech toolkits.

39

COLLABORATION and COOPERATION TTIC expanded its efforts to engage in future research, academic coordination and exchange when it entered into an initial agreement during 2015-16 with the School of Computing, Tokyo Institute of Technology, and signed a Letter of Intent with the National Institute of Advanced Industrial Science and Technology Artificial Intelligence Research Center in Japan. In both instances, the institutions express a desire to work together in the future and collaborate to engage researchers and students for academic exchange if the opportunity arises, and to promote partnerships. TTIC was one of fourteen institutions, departments and organizations that sponsored the 2015 Linguistic Summer Institute held in July 2015 at the University of Chicago. The four-week program was intended for anyone engaged in the academic discipline of linguistics: undergraduates, graduate students, and faculty members from countries around the world. TTIC’s Prof. Karen Livescu was a course instructor for Speech Technologies at the Summer Institute and her course introduced techniques used in speech technologies, mainly focusing on automatic speech recognition (ASR). Mohit Bansal collaborated with Profs. Dhruv Batra and Devi Parikh of Virginia Tech and their students, some of whom interned with Prof. Bansal at TTIC. They worked on humor prediction and generation, visual story sorting, and visual question answering. Prof. Bansal also collaborated with Makoto Miwa of TTI Japan on joint entity and relation extraction neural models, and with IBM Research members David McClosky (now at Google Research) and Sid Patwardhan, and a BarIlan intern on the role of context types and dimensionality in learning word embeddings. Srinadh Bhojanapalli has ongoing collaboration with Prateek Jain and Praneeth Netrapalli at Microsoft Research India on various high dimensional statistical learning problems. He has started new collaboration on algorithms for smooth non-convex optimization with Stephen Wright, a TTIC adjoint professor from the University of Wisconsin. Stefan Canzar is currently collaborating with a number of research groups in ongoing projects, such as with Jef Boeke of NYU School of Medicine and Zheng Kuang of the University of Texas Southwestern Medical Center studying alternative splicing during meiosis in fission yeast. He has also developed a metric to compare phylogenetic trees with Domagoj Matijević of the

40

University of Osijek and Khaled Elbassioni of the Masdar Institute of Science and Technology. Along with Hongjun Son of Johns Hopkins Medicine he has studied the regulation of translation in neurons. Natalia Maltsev from the Department of Human Genetics, University of Chicago, and Prof. Canzar have been working on an automatic pipeline for the functional annotation of isoforms. He has also been in collaboration with Heejung Shim of the Department of Statistics at Purdue University. They are developing a probabilistic model for alternative splicing. Ayan Chakrabarti continued his collaboration with Kalyan Sunkavalli at Adobe's Imagination lab, and started a new collaboration with Prof. Sanjeev Koppal at the University of Florida. Kevin Gimpel collaborated with Hua He and Jimmy Lin at the University of Maryland at College Park on deep learning for natural language processing. He is also involved in an ongoing collaboration with Aynaz Taheri and Tanya Berger-Wolf at the University of Illinois at Chicago related to using neural networks to learn graph representations. Hammad Naveed collaborated with Jie Liang of University of Illinois at Chicago on computational modeling of beta barrel membrane proteins. Prof. Naveed also collaborated with Dr. Michael Maitland of University of Chicago Medicine on discovering novel targets for FDA-approved kinase inhibitors. Prof. Naveed also collaborated with Drs. Xin Gao and Stefan Arold of King Abdullah University of Science and Technology, Saudi Arabia, on computational characterization of protein drug interactions. Matthew Walter has an ongoing collaboration with several universities, including the University of Rochester, Carnegie Mellon University, Massachusetts Institute of Technology, University of Pennsylvania, University of Central Florida; and institutions such as NASA's Jet Propulsion Laboratory and the Army Research Lab, as part of the Robotics Collaborative Technology Alliance (RCTA) to develop mobile robots and mobile manipulators that work alongside soldiers. The project is funded by the U.S. Army. Prof. Walter also has a collaboration with MIT Lincoln Laboratory as part of an Office of Naval Research project focused on the use of heterogeneous teams of robots (wheeled and aerial vehicles) for mapping and surveillance.

41

TALKS, SEMINARS AND WORKSHOPS Talks and seminars are an important part of any academic institution. They are both a way for researchers to promote their research, and to keep abreast of recent developments. They play an important role in establishing the level of intellectual activity and influx of innovative ideas at an organization: research is more likely to be productive in an active environment with significant interaction between researchers. The table below lists seminars given at TTIC, many of which are given by speakers from other universities and research institutions, as part of the TTIC Colloquium: a forum for talks by invited speakers on work of current relevance and broad interest to the computer science community. Other talks may be a part of the Research

at TTIC series: a weekly seminar series presenting research currently underway at the institute. Every week a different TTIC faculty member will present their research. The lectures are intended both for students seeking research topics and advisors, and for the general TTIC and University of Chicago communities interested in hearing what their colleagues are currently involved. New for 2015-16 is the Young Researcher Seminar

Series featuring talks by PhD students and postdocs whose research is of broad interest to the computer science community. The series provides an opportunity for early-career researchers to present recent and promising work and to meet with students and faculty at TTIC and nearby universities. Lastly, some speakers may be part of research Reading Groups: people presenting papers that are of interest to a particular group, such as the theory group or the programming languages group. Most seminars are advertised outside of TTIC and are intended to be for a broad audience in computer science. In the spring quarter there are a large number of recruiting seminars which are talks given by candidates for faculty positions. The TTIC Event Calendar can be found via the TTIC homepage: www.ttic.edu

Speaker Yuan Yao Sharon Goldwater

Paul Smolensky

42

Institute Peking University University of Edinburgh Johns Hopkins University

Title

Date

Sparse Recovery via Dynamics

7/15/15

Unsupervised Word Segmentation and Lexicon Discovery from Speech using Acoustic Word Embeddings

7/17/15

What can vectorial representations do? Distributed structure processing in neural networks

7/20/15

Jacob Steinhardt

Stanford University

Learning with Intractable Inference and Partial Supervision

7/27/15

Takeshi Onishi

TTIC

Student Talk: Neural Coreference Resolution

7/28/15

Tomer Michaeli

Technion

Blind deblurring and blind super-resolution using internal patch recurrence

8/3/15

TTIC

Student Talk: Bounded Complexity Families and Isomorphism Conjecture

8/10/15

Approximate Matchings in Dynamic Graph Streams

8/10/15

Sampling Strategies for Feature-Efficient and Active Learning

8/17/15

Weizmann Institute of Science

Vertex Sparsification of Cuts, Flows, and Distances

8/31/15

Slav Petrov

Google Research NYC

Towards Universal Syntactic Processing of Natural Language

9/4/15

Jianzhu Ma

TTIC PhD Candidate

Thesis Defense: Protein Structure Prediction by Protein Alignments

9/17/15

Machine Learning Approaches to Speech Enhancement

9/28/15

Random Walks on Discourse Spaces: a Generative Model Approach to Semantic Word Embeddings

9/30/15

Fast, Approximate and Scalable Geometric Optimization

10/5/15

Improved 3d Structure Prediction of Beta-Barrel Membrane Proteins Using Evolutionary Coupling Constraints and a Reduced State Space

10/9/15

A Big World of Tiny Motions

10/15/15

Mrinalkanti Ghosh Sanjeev Khanna Lev Reyzin Robert Krauthgamer

Paris Smaragdis

Sanjeev Arora

Chandrajit Bajaj Hammad Naveed

William T. Freeman

University of Pennsylvania University of Illinois, Chicago

University of Illinois, UrbanaChampaign Princeton University University of Texas at Austin TTIC

Massachusetts Institute of Technology

Nati Srebro

TTIC

Learning and Optimization: Deep and Distributed

10/16/15

Srinadh Bhojanapalli

TTIC

Dropping Convexity for Faster Semi-definite Optimization

10/23/15

Huy Nguyen

TTIC

Distributed Machine Learning

10/30/15

Diderot: a Domain-Specific Language for Portable Parallel Scientific Visualization and Image Analysis

11/2/15

Gordon Kindlmann

University of Chicago

43

Li-Yang Tan

TTIC

An Average-Case Depth Hierarchy Theorem for Boolean Circuits

11/6/15

Tara Sainath

Google Research

Single and Multichannel Raw Waveform Neural Network Acoustic Models

11/9/15

TTIC

Rich Representations for Parsing Visual Scenes

11/13/15

Greg Shakhnarovich Matthew Stephens

University of Chicago

False Discovery Rates - a new deal

11/16/15

Lihi Zelnik-Manor

Technion / Cornell Tech

Approximate Nearest Neighbor Methods in Computer Vision, Beyond L2

11/23/15

University of Illinois at Chicago

Analysis of Dynamic Interaction Networks

12/14/15

Tanya Berger-Wolf Qixing Huang

TTIC

Gunnar Klau

Centrum Wiskunde

Mehrdad Mahdavi John Wright

TTIC Carnegie Mellon University

1/8/16

Mining Gummi Bears in One and Two Jars

1/11/16

Stochastic optimization with exponentially concave losses: lower and upper bounds on the excess risk

1/15/16

Random Words, Longest Increasing Subsequences, and Quantum PCA

1/18/16

Affinity CNN: Learning Pixel-Centric Pairwise Relations for Figure/Ground Embedding

1/22/16

Michael Maire

TTIC

Zhiyong Wang

TTIC PhD Candidate

Thesis Defense: Knowledge-Based Machine Learning Methods for Macromolecular 3D Structure Prediction

1/25/16

Dani Yogatama

Carnegie Mellon University

Learning to Represent Language: Embeddings and Optimization

1/25/16

Analyzing Complex Systems and Networks: Incremental Optimization and Robustness

1/27/16

Architectures for Learning in Low-level Vision Applications

1/29/16

Mert Gurbuzbalaban

Ayan Chakrabarti

44

& Informatica, Amsterdam

Dense Correspondences in the Era of Deep Learning

Massachusetts Institute of Technology TTIC

Bundit Laekhanukit

The Weizmann Institute of Science

Directed Network Design and Related Problems

2/1/16

Yajie Miao

Carnegie Mellon University

End-to-End Speech Recognition using Deep LSTMs, CTC Training and WFST Decoding

2/3/16

Boxin Shi

Nanyang Technological University

Camera Intelligence from Visual Computation and Sensor Innovation

2/4/16

Karen Livescu Xiaorui Sun

TTIC Columbia University

Toward Neural Segmental Sequence Models

2/5/16

Efficient Density Estimation via Piecewise Polynomial Approximation

2/8/16

Ilya Razenshteyn

Massachusetts Institute of Technology CSAIL

Locality-Sensitive Hashing and Beyond

2/10/16

Tandy Warnow

University of Illinois at UrbanaChampaign

New HMM-Based Methods in Sequence Alignment, Phylogenetics, and Metagenomics

2/12/16

Machine Learning for Observational Studies

2/17/16

Mitsubishi Electric Research Laboratories (MERL)

Towards Holistic Scene Understanding from a Single Image

2/18/16

University of Texas at Austin

Mining Structured Matrices in High Dimensions

2/19/16

Using Visual and Linguistic Information to Connect Language to the World

2/22/16

Ben-Gurion University and Harvard University

Differential Privacy for Data Analysis

2/23/16

Olga Russakovsky

Carnegie Mellon University

The Human Side of Computer Vision

2/24/16

Iman Hajirasouliha

Stanford University

Computational methods for characterizing largescale human genome variations with applications to cancer

2/24/16

Fast and Simple PCA via Convex Optimization

2/26/16

Topological Sampling Problems

2/29/16

Using Motion to Understand Objects in the Real World

3/2/16

Algorithms to Bring Realistic Modeling to Drug Design

3/4/16

University of Washington

Interactive Machine Learning for Information Extraction

3/7/16

Stanford University

Initialization and Dual Expressivity of Neural Networks

3/9/16

Reed-Solomon Codes: from Theory to Practice.

3/10/16

Uri Shalit Srikumar Ramalingam

Suriya Gunasekar Carina Silberer Kobbi Nissim

Dan Garber Shmuel Weinberger

New York University

University of Edinburgh

TTIC University of Chicago

David Held

Stanford University

Mark Hallen

Duke University

Sameer Singh Roy Frostig Mary Wootters

Carnegie Mellon University

45

Liang Lu

Deep Learning for End-to-End Speech Recognition

3/11/16

Mark Schmidt

University of British Columbia

Advances in Solving Structured Optimization Problems.

3/14/16

S. Matthew Weinberg

Massachusetts Institute of Technology

Algorithms for Strategic Agents

3/16/16

Towards Deep Convolutional Methods without Supervision

3/28/16

Johns Hopkins University

Compute Faster and Learn Better: Machine Learning via Nonconvex Model-based Optimization

3/30/16

Dhruv Batra

Virginia Tech

Towards Transparent Intelligent Systems: Diverse Predictions from Perception Modules

3/31/16

Devi Parikh

Virginia Tech

Words, Pictures, and Common Sense

4/1/16

TTIC

Learning Concise Representations of Textual Knowledge

4/8/16

Andreas Spanias

Arizona State University

An Introduction on the Activities at the SenSIP Center.

4/8/16

YangFeng Ji

Georgia Tech

Distributed Representation Learning for Discourse Processing

4/11/16

William Wang

Carnegie Mellon University

Scalable Learning and Reasoning for Large Knowledge Graphs

4/13/16

Learning Image Representations from Unlabeled Video

4/18/16

Neural Dialogue Generation

4/20/16

Architectural Complexity Measures of Recurrent Neural Networks

4/21/16

TTIC

Satisfiability of Ordering CSPs Above Average

4/22/16

Carnegie Mellon University

Learning Large Graphs from Compressed and Subsampled Data

4/27/16

Thesis Defense: American Sign Language Fingerspelling Recognition from Video: Methods for Unrestricted Recognition and Signer-Independence

4/29/16

Handling and Harnessing Data Scarcity

4/29/16

Cognitive and Application-Driven Machine Learning for Natural Language

5/6/16

Zaid Harchaoui Tuo Zhao

Kevin Gimpel

Kristen Grauman Jiwei Li Yuhuai Wu Yury Makarychev Gautam Dasarathy Taehwan Kim

Mesrob Ohannessian

Christos Christodoulopoulos

46

University of Edinburgh

New York University

University of Texas, Austin Stanford University University of Toronto

TTIC PhD Candidate University of California, San Diego University of Illinois at UrbanaChampaign

Saurabh Gupta

Ofer Meshi

University of California, Berkeley TTIC

Scene Understanding from RGB-D Images

5/11/16

Optimization and tightness of convex relaxations for structured output prediction

5/13/16

Sanjeev J. Koppal

University of Florida

Privacy Preserving Optics for Miniature Vision Sensors

5/16/16

Shayan Oveis Charan

University of Washington

Applications of strongly Rayleigh distributions in Algorithm Design

5/19/16

TTIC

On the Approximability of Constraint Satisfaction Problems

5/20/16

Hebrew University

High-dimensional Permutations and Discrepancy

5/25/16

Complexity of Semi-Algebraic and Algebraic Proofs

5/27/16

From Probabilistic Models to Decision Theory and Back Again

6/2/16

Dense Correspondences in the Era of Deep Learning

6/3/16

Madhur Tulsiani Nati Linial Alexander Razborov

TTIC

Sanmi Koyejo

Stanford University

Qixing Huang

TTIC

Brian Ziebart

University of Illinois, Chicago

Supervised Machine Learning as an Adversarial Game

6/6/16

John Lafferty

University of Chicago, TTIC

The Complexity of Minimizing Individual Convex Functions

6/10/16

Rayid Ghani

University of Chicago

Machine learning for Public Policy and Social Good: Case Studies, Challenges, and Opportunities

6/13/16

Token Embeddings: Embedding Words in Context for Syntactic Tasks

6/27/16

Lifu Tu

TTIC

47

Distinguished Lecture Series 2015-16 September 2015 through February 2016. (Location: TTIC) Speakers included: Sanjeev Arora Charles C. Fitzmorris Professor of Computer Science, Princeton University. Talk Title: Random walks on discourse spaces: A generative model approach to semantic word embeddings. William T. Freeman Professor of Electrical Engineering and Computer Science, Member of Computer Science and Artificial Intelligence Laboratory, Massachusetts Institute of Technology. Talk Title: A big world of tiny motions. Yoshua Bengio Professor of Computer Science and Operations Research, Canada Research Chair in Statistical Learning Algorithms, Universite de Montreal. Talk Title: Deep Learning Theory. Tandy Warnow Professor of Computer Science and Bioengineering, The University of Illinois at Urbana-Champaign. Talk Title: New HMM-based Methods in Sequence Alignment, Phylogenetics, and Metagenomics.

First Annual Midwest Robotics Workshop at TTIC On March 17-18, 2016, TTIC hosted the first annual Midwest Robotics Workshop (MWRW). The workshop is intended to bring together roboticists from academia and industry in and around the Midwestern United States. It is an opportunity for researchers and practitioners to share their work and to network with one another, with the goal of creating a more cohesive and vibrant robotics community in the Midwest. The workshop featured invited talks by leading researchers, and an exciting collection of oral presentations and interactive poster sessions. Workshop website: http://www.ttic.edu/mwrw/

48

Midwest Vision Workshop On April 14-15, 2016, TTIC once again hosted the Midwest Vision Workshop. This is a regular regional meeting of computer vision researchers, providing a forum for presenting recent work, informal discussion and exchange of ideas. The meeting, which included oral presentations and poster sessions, draws participants from TTIC, University of Chicago, University of Illinois at Urbana-Champaign, University of Michigan at Ann Arbor, Indiana University, Michigan State, Washington University in St. Louis, and other institutions. The two-day program included talks and poster presentations.

49

EDUCATION The PhD Program The TTIC PhD Program is designed to prepare students for academic or research careers in computer science. To complete the program, a student must make an original and significant contribution to the field of computer science and this contribution must be described in a doctoral thesis. In addition to the thesis, there are course and examination requirements to complete the program. The main component of the program is the process by which the student learns to do research and becomes a part of the academic community. As part of the associated partnership between TTIC and the University of Chicago, students of TTIC can take and receive credit for courses through the University of Chicago, and University of Chicago students can take advantage of classes TTIC offers as well. Students have taken full advantage of this opportunity. TTIC’s students have full access to the University of Chicago library system, athletic facilities, the student health center and transportation on campus. They may enjoy the benefits and great rewards of an intimate learning, study and research setting, exposure to state-of-the-art research, opportunities in the greater computer science community, and a shared and traditional experience that come with association with a large university.

50

Student Progress TTIC student Jianzhu Ma, studying under Professor Jinbo Xu, successfully defended his theses this year and will receive his doctoral diploma in a ceremony in September 2016. This will be the seventh doctoral diploma to be awarded. Jianzhu Ma’s research interest is in bioinformatics. He is currently employed at the University of California, San Diego, Department of Medicine. TTIC has several more PhD Candidates working on their theses and expects two more graduates by the end of summer 2016 to also participate in the September 2016 diploma ceremony. Students Mrinalkanti Ghosh, Mohammadreza Mostajabi, Qingming Tang and Shubhendu Trivedi successfully fulfilled all requirements to complete the Master’s portion of the PhD Program, and received master’s diplomas from the Institute at the September 2015 diploma ceremony at the start of the academic year.

51

TTIC Curriculum Serving the University of Chicago TTIC instructors serve the TTIC student population in in their courses, and under the TTIC-University of Chicago Agreement, University students may enroll in TTIC’s courses and receive credit through the University, and vice-versa. TTIC views this as part of serving the Education Mission of the Institute. The amount of University students who register for TTIC courses has been increasing the last several years. Course TTIC 31020: Introduction to Statistical Machine Learning, taught by Prof. Greg Shakhnarovich, had thirty-three attendees this Autumn Quarter 2015. Sixty-five percent of the course attendees were University of Chicago students. This is up from twenty-four registered students in the course the year before. TTIC instructors are proud to offer a quality curriculum to its PhD students and share knowledge with the quality students from the University who enroll in Institute courses.

TTIC Student Awarded University of Chicago TA Prize As part of TTIC and the University of Chicago’s agreements of engagement, both institutions share courses, services, and collaborate in many areas. As part of this, it is not uncommon for a TTIC student to assist with TAing a University course, or a University student TAing a TTIC course. At the University annually, students are asked who they believe deserves the Department of Computer Science’s Teaching Assistant prize. These TAs stand out for dedication to students, and go above and beyond to help students gain an understanding of the concepts covered by a course. TTIC PhD Candidate Shubhendu Trivedi was awarded the TA Prize for 2016 for his TA work in the course CMSC 25400/STAT 27725: Machine Learning. TTIC is very proud of Shubhendu’s dedication and talents in the classroom, and for his service to the greater campus community outside of TTIC.

52

Student Publications, Posters and Abstracts Tang, Qingming, Chao Yang, Jian Peng and Jinbo Xu. “Exact Hybrid Covariance Thresholding for Joint Graphical Lasso.” Paper presented at European Conference on Machine Learning and Knowledge Discovery in Databases (ECML-PKDD), Porto, Portugal, August 2015. doi:10.1007/978-3-319-23525-7_36. Tang, Qingming, Sheng Wang, Jian Peng, Jianzhu Ma, and Jinbo Xu. “Bermuda: De Novo Assembly of Transcripts with New Insights for Handling Uneven Coverage.” Paper presented at the ACM Conference on Bioinformatics, Computational Biology and Health Informatics, Atlanta, GA, September 2015. doi:10.1145/2808719.2808736. Hashemifar, Somaye, Behnam Neyshabur, and Jinbo Xu. "Joint Inference of Tissue-specific Networks with a Scale Free Topology." Paper presented at IEEE International Conference on Bioinformatics and Biomedicine (IEEE BIBM), Washington, DC, November 2015. doi:10.1109/BIBM.2015.7359696. Sun, Siqi, Jianzhu Ma, Sheng Wang, and Jinbo Xu. “Predicting Diverse M-best Protein Contact Maps.” Paper presented at the IEEE International Conference on Bioinformatics and Biomedicine (BIBM), Washington, DC, November 2015. doi:10.1109/BIBM.2015.7359865. Neyshabur, Behnam, Ruslan Salakhutdinov, and Nathan Srebro. "Path-SGD: Path-Normalized Optimization in Deep Neural Networks." Paper presented at Neural Information Processing Systems (NIPS), Montreal, December 2015. arXiv:1506.02617. Wang, Sheng, Jian Peng, Jianzhu Ma, and Jinbo Xu. “Protein Secondary Structure Prediction Using Deep Convolutional Neural Fields.” Scientific Reports 6 (January 2016). doi:10.1038/srep18962. Kim, Taehwan, Weiran Wang, Hao Tang, and Karen Livescu. “Signer-independent Fingerspelling Recognition with Deep Neural Network Adaptation.” Paper presented at International Conference on Acoustics, Speech, and Signal Processing (ICASSP), Shanghai, China, March 2016. doi:10.1109/ICASSP.2016.7472861. Liu, Chunxi, Preethi Jyothi, Hao Tang, Vimal Manohar, Rose Sloan, Tyler Kekona, Mark Hasegawa-Johnson, and Sanjeev Khudanpur. “Adapting ASR for Under-resourced Languages Using Mismatched Transcriptions.” Paper presented at In International Conference on Acoustics, Speech, and Signal Processing (ICASSP), Shanghai, China, March 2016. doi:10.1109/ICASSP.2016.7472797. Hashemifar, Somaye, Qixing Huang, and Jinbo Xu. “Joint Alignment of Multiple Protein-Protein Interaction Networks via Convex Optimization.” Paper presented at International Conference on Research in Computational Molecular Biology (RECOMB), Los Angeles, CA, April 2016. doi:10.1089/cmb.2016.0025. Choi, Heejin, Ofer Meshi, and Nathan Srebro. "Fast and Scalable Structural SVM with Slack Rescaling." Paper presented at International Conference on Artificial Intelligence and Statistics (AISTATS), Cadiz, Spain, May 2016. arXiv:1510.06002v2. Neyshabur, Behnam, Ryota Tomioka, Ruslan Salakhutdinov, and Nathan Srebro. "Data-Dependent Path Normalization in Neural Networks." Paper presented at International Conference on Learning Representations (ICLR), San Juan, Puerto Rico, May 2016. arXiv:1511.06747v4. Wieting, John, Mohit Bansal, Kevin Gimpel, and Karen Livescu. “Toward Universal Paraphrastic Sentence Embeddings.” Paper presented at International Conference on Learning Representations (ICLR), San Juan, Puerto Rico, May 2016. arXiv:1511.08198v3. He, Hua, John Wieting, Kevin Gimpel, Jinfeng Rao, and Jimmy Lin. “UMD-TTIC-UW at SemEval-2016 Task 1:

53

Attention-Based Multi-Perspective Convolutional Neural Networks On Textual Similarity Measurement.” Paper presented at SemEval, San Diego, CA, June 2016. Li, Xiang, Aynaz Taheri, Lifu Tu, and Kevin Gimpel. "Commonsense Knowledge Base Completion." Paper presented at Association for Computational Linguistics (ACL), Berlin, August 2016. doi:10.18653/v1/ P16-1137. Kveton, Branislav, Hung Bui, Mohammad Ghavamzadeh, Georgios Theocharous, S. Muthukrishnan and Siqi Sun. “Graphical Model Sketch.” Paper to be presented at European Conference on Machine Learning and Knowledge Discovery in Databases (ECML-PKDD), Riva Del Garda, Italy, September 2016. doi:10.1007/978-3-319-46128-1_6. Tang, Hao, Weiran Wang, Kevin Gimpel, and Karen Livescu. “Efficient Segmental Cascades for Speech Recognition.” Paper to be presented at Interspeech, San Francisco, CA, September 2016. doi:10.21437/Interspeech.2016-1298. Wang, Weiran, Hao Tang, and Karen Livescu. “Triphone State-tying via Deep Canonical Correlation Analysis.” Paper to be presented at Interspeech, San Francisco, CA, September 2016. doi:10.21437/ Interspeech.2016-1300. Wang, Sheng, Siqi Sun and Jinbo Xu. “AUC-maximized Deep Convolutional Neural Fields for Sequence Labeling.” Paper to be presented at European Conference on Machine Learning and Knowledge Discovery in Databases (ECML-PKDD), Riva Del Garda, Italy, September 2016. doi:10.1007/978-3-319 -46227-1_1. Sulakhe, Dinanath, Bingqing Xie, Andrew Taylor, Mark D’Souza, Sandhya Balasubramanian, Somaye Hashemifar, Steven White, Utpal J. Dave, Gady Agam, Jinbo Xu et al. “Lynx: A Knowledge Base and an Analytical Workbench for Integrative Medicine.” Nucleic Acids Research 44, no. D1 (November 2016): D882--D887. doi:10.1093/nar/gkv1257. Wieting, John, Mohit Bansal, Kevin Gimpel, and Karen Livescu. “Charagram: Embedding Words and Sentences via Character n-grams.” Paper presented at Empirical Methods in Natural Language Processing (EMNLP), Austin, November 2016. arXiv:1607.02789. Tang, Hao, Weiran Wang, Kevin Gimpel, and Karen Livescu. “End-to-End Training Approaches for Discriminative Segmental Models.” Paper presented at IEEE Workshop on Spoken Language Technology (SLT), San Diego, CA, December 2016. arXiv:1610.06700. Toshniwal, Shubham, and Karen Livescu. "Jointly Learning to Align and Convert Graphemes to Phonemes with Neural Attention Models." Paper to be presented at IEEE Workshop on Spoken Language Technology (SLT), San Diego, CA, December 2016. arXiv:1610.06540. Woodworth, Blake, and Nathan Srebro. "Tight Complexity Bounds for Optimizing Composite Objectives." Paper presented at Conference on Neural Information Processing Systems (NIPS), Barcelona, Spain, December 2016. arXiv:1605.08003v2. Tang, Qingming, Lifu Tu, Weiran Wang, and Jinbo Xu. “Network Inference by Learned Node-Specific Degree Prior.” Preprint, submitted February 7, 2016. arXiv:1602.02386.

54

TTIC’s Largest Incoming Class In the fall of 2004, TTIC matriculated its first three students. The 2015-16 academic year began with twenty-eight students, nine who were newly enrolled for Autumn 2015: the Institute’s largest incoming class to date. Four students have been admitted and will enroll for the 2016-17 year.

Financial Support for Students Full financial support is offered to all enrolled students in good standing, making progress towards their degree, guaranteed for four years. The tuition for an academic year is $30,000. All students at TTIC may expect to receive financial support that covers tuition, health services, health insurance and student life fees, and a scholarship to assist with living expenses, provided they remain full-time and in good academic standing.

Exchange Students This year TTIC welcomed two exchange students from the Toyota Technological Institute located in Nagoya, Japan (TTIJ). Tomoaki Kondo and Chenxi Yang arrived in September 2015, took TTIC and University of Chicago courses, and returned to TTIJ in late December. TTIC remains pleased with the exchange program with TTIJ, as the experience continues to be a positive success for all involved. Two new exchange students are scheduled to enroll on exchange at the institute for Autumn Quarter 2016.

55

INSTITUTE GOALS Accreditation On November 5, 2015, the Higher Learning Commission (HLC) Board reaffirmed TTIC's Accreditation, placing the Institute on Standard Pathway. The reaffirmation letter of notice requires that TTIC: • • •

submit an October 2017 Interim Report to the Commission (marking progress on several key points), apply for a Substantive Change Application for the TTIC Master’s degree, and then be prepared for the normal 2019-20 four-year comprehensive visit, and the 2025-26 reaffirmation visit.

TTIC was pleased with this outcome, which came less than two months after TTIC attended a mandatory hearing with the Commission’s Institutional Actions Council (IAC) regarding the the Commission’s on-site visit to TTIC from November 2014. The September 2015 hearing covered topics of Board autonomy, financial independence, assessment and student retention and completion. TTIC is thankful to the HLC for its partnership in working together to assist the institute, and the internal and external constituents who gave their time, support and energy to the accreditation process. TTIC is improved because of it. In December 2015, TTIC submitted a Substantive Change Application to the HLC. As TTIC only admits PhD students, the PhD program as a whole was accredited. Until this point, this included the Master’s within the PhD portion, inside the program, but Master’s-within-the-PhD diplomas/degrees were somewhat of a gray area. The commission wanted a full review of the diploma requirements to determine if it met standards. In February 2016, TTIC was notified that there would be a Change Visit. The Change Visit was held June 20-21, 2016, at TTIC. Two HLC agents came to TTIC and met with leadership, faculty, students and administration, as well as Institute Trustees. They reviewed all requirements for the Master’s diploma. The agents have reported to TTIC that they will be recommending to the HLC Institutional Actions Council that TTIC be recognized to award the Master’s degree. That determination will be made in August 2016.

56

Board Strategy for Continued Improvements The Board appointed three new distinguished Trustees in July of 2015: • • •

Dr. Eric Grimson, Chancellor for Academic Advancement, Massachusetts Institute of Technology; Executive Advisory Committee, TTIC Dr. Jim Merz, Frank M. Freimann Professor Emeritus of Engineering, Concurrent Professor of Physics, University of Notre Dame Dr. Richard Samuels, Ford International Professor, Dept. of Political Science, Massachusetts Institute of Technology; Director, MIT Center for International Studies; Founding Director, MIT Japan Program

The new Trustees are outstanding academic appointees, independent, and were selected via processes carried out by the Board Search Committee. Changes were also approved to the Bylaws regarding the description of two Officer positions: Treasurer and Secretary to the Board, corresponding with Article 5, Sections 5.7 and 5.8. Changes included: • • • • • • •

The Treasurer’s reporting of investment will now be a responsibility of the Chairperson(s) of the Finance Committee. The Chief Financial Officer will be a staff administrator who reports day-to-day to the President, and acts under the direction of the Board; the CFO must be independent of any major donor. The CFO will invest funds at the direction of the President and the Board. TTIC will now have a President, Chief Financial Officer (CFO) and a Secretary of the Institute as Officers. The Board Chair will no longer be an Officer. Secretary of the Board becomes a Secretary of the Institute, falling under administration, supporting governance. The Secretary and CFO positions are now under a three-year appointment.

Committees received attention in 2015-16 as well. A new committee of the Board was created: the President’s Tenure Appointment Review Committee (PTAR). The committee shall review cases for faculty promotion and tenure, and make final recommendations to the President. It was approved by the full Board that committee service terms shall be three years, with an option for renewal. An independent majority of membership was also expanded in the Executive Committee. The Board received its first TRIP (Total Return Investment Pool) Investment Report from the University of Chicago Investment Office since its initial endowment investment with the office. The Institute will be receiving quarterly reports., Report review will be a function of the Chief Financial Officer and the Finance Committee, reporting to the Board. See p. 59 section Endowment Growth and Investment Collaboration with University of Chicago .

57

External Advisory Committee Visit 2016 TTIC has been utilizing the External Advisory Committee (EAC) as a resource since the fall of 2004. This committee met on the Institute’s behalf once more in February 2016. The EAC was on-site February 15-16, met with institute leadership, discussed a tenure case with senior faculty including adjoint faculty, held a session with current students, and also held a session with current tenured and tenure-track faculty. The current EAC membership includes: Eric Grimson, Chancellor for Academic Advancement, Massachusetts Institute of Technology Takeo Kanade, UA and Helen Whitaker Professor, Carnegie Mellon University Richard Karp, Professor of Electrical Engineering and Computer Science, University of California, Berkeley Éva Tardos, Jacob Gould Schurman Professor of Computer Science, Cornell University Based on the visit, and supporting material supplied to the committee, the committee provided TTIC with a final report that outlined the EAC’s impression of TTIC’s state of progress, any areas of concern, and their general observations. Their report noted areas of progress that that the Institute has pursued: accreditation efforts, long-term financial stability, strong and committed service of the President, and a better-utilization of senior faculty. The committee then made some recommendations for the Institute which touched on areas of expanding external funding, intellectual property policies, nurturing the growth of senior faculty and their service to the Institute and involvement in oversight, engagement of external faculty contributors, promoting and acting assertively on a strategy for gender diversification in faculty and student recruitment, student satisfaction, University of Chicago interaction at the student level, and the strategic plan of the institute. The Committee also discussed the transition of Chief Academic Officer and the exposure of faculty. The report and guidance that the EAC provides to TTIC is invaluable. Their recommendations serve to direct TTIC in best practices regarding operational improvements, service to faculty and students, and quality insistence in all aspects. We thank them for their continued service on the committee and on behalf of the Institute’s mission and future.

58

Endowment Growth and Investment Collaboration with University of Chicago In June 2015, the Toyota Motor Corporation (TMC) provided a donor agreement to TTIC stating: TMC shall donate the sum of $85 million to TTIC's endowment by the following three installations; 1st installment: $30 million by the end of June, 2015; 2nd installment: $30 million by the end of July, 2016; 3rd installment: $25 million by the end of July, 2017. As approved by the Board of Trustees, in July of 2015, TTIC and the University of Chicago entered into an Agreement of Investment Management, which outlines the terms of investment in the University’s Total Return Investment Pool (TRIP) in which TTIC is investing $50 million of its endowment. The agreement was signed by President Sadaoki Furui and University of Chicago Vice President and Chief Investment Officer, Mark Shmid. The agreement remains valid until TTIC cashes out all TRIP funds, or with 180 days written notice by the Investment Office. The Board also approved further investment into TRIP by authorizing the investment of a portion of the $30 million contribution from Toyota Motor Corporation that will be received in July of 2016.

Computing Capacity Doubles in 2016 TTIC doubled its computing capacity in 2016. There now are sixty state-of-the-art GPUs (Graphics Processing Units) with a total of 684 Gigabytes of GPU memory and 520 CPU (Central Processing Unit) cores with a total of 4,416 Gigabytes of system memory. All of these computing resources are served by a total of 150 Terabytes of shared storage.

TTIC’s EAC members: (From top left to bottom right) Richard Karp, Eric Grimson, Eva Tardos, Takeo Kanade.

59

INTERNS AND VISITING SCHOLARS TTIC maintains a steady number of interns and visiting scholars who engage in study and research on the premises. There were twenty-seven visiting scholars from other institutions in the U.S. and abroad who came to the Institute to work with TTIC faculty. These short-term visiting scholars bring interest, energy, and enthusiasm to our academic community, and allow TTIC students access to a broad range of specialties that outside researchers bring with them, along with ideas and culture brought from the visitors’ home institutions.

Deblin Bagchi, Ohio State University (K. Livescu)

Dian Li, University of Chicago (M. Bansal, M. Walter)

Vijay Bhattiprolu, Carnegie Mellon University (M. Tulsiani)

Zhen Li, University of Hong Kong (J. Xu)

Luka Borozan, University of Osijek, Croatia (S. Canzar) Bharat Chandar, University of Chicago (M. Walter) Arjun Chandrasekaran, Virginia Tech (M. Bansal) Cha Chen, University of Chicago (M. Bansal, M. Walter) Andrea F. Daniele, Sapienza University of Rome (M. Bansal, M. Walter) Alex Gajewski, University of Chicago Lab School (M. Walter) Arnab Ghosh, Indian Institute of Technology Kanpur (M. Bansal) Wanjia He, University of Chicago (K. Livescu) Dan Hendrycks, University of Chicago (K. Gimpel) Dong Ki Kim, Cornell University (M. Walter) Myungin Kim, University of Chicago (M. Bansal) Aravind Srinivas Lakshminarayanan, Indian Institute of Technology Madras (M. Bansal)

60

Raci Lynch, Stanford University (K. Livescu) Vivek Madan, University of Illinois, Urbana-Champaign (J. Chuzhoy) Sepideh Mahabadi, Massachusetts Institute of Technology (J. Chuzhoy, Y. Makarychev) Pasin Manurangsi, University of California, Berkeley (Y. Makarychev, M. Tulsiani) Yixin Nie, University of Chicago (M. Bansal) Manasvi Sagarkar, University of Chicago (K. Gimpel) Ridwan Syed, University of Chicago (M. Tulsiani) Trang Tran, University of Washington (M. Bansal, K. Gimpel, K. Livescu) Igor Vasiljevic, University of Chicago (G. Shakhnarovich) Yue Xu, California State University (J. Xu) Xiaoming Zhao, University of Science and Technology of China (J. Xu)

Research Highlights The Institute’s visiting students are working in a diverse range of research areas, including theory, machine learning, robotics, natural language processing, computer vision, computational biology, and speech technologies. Below are highlights of research projects with profiles of a few of the visiting students:

Andrea Daniele is a visiting student from the University of Rome, La Sapienza, where he has been studying Artificial Intelligence and Robotics Engineering. At TTIC, he is working with Matthew Walter and Mohit Bansal on natural language generation in the context of providing indoor route instructions, which is the problem of planning and synthesizing route instructions with the objective to allow people to easily navigate unknown environments. Andrea is broadly interested in research topics involving both NLP and Human-Robot interaction. He is from Petilia Policastro, a small village in Italy and likes to listen to Jazz and visit museums in his free time.

Dan Hendrycks is an undergraduate at the University of Chicago studying computer science. At TTIC he is working with Kevin Gimpel on the architecture of neural networks and practical deep learning security, including recognizing adversarial images and speech. Dan is from the rural town of Marshfield, Missouri, and enjoys advocating for various effective causes.

Sepideh Mahabadi is a PhD student at the theory of computation group at CSAIL, MIT. Her advisor is Piotr Indyk and her research interests mainly include high dimensional geometry, streaming algorithms, and graph algorithms. She is doing an internship at TTIC working with Julia Chuzhoy and Yury Makarychev on graph algorithms and embedding. She is from Tehran and did her undergraduate studies at Sharif University, Iran.

Trang Tran is a visiting student from the University of Washington, where she works on various speech and language processing projects. Her research ranges from modeling the language of online discussions to studying acoustic-prosodic features indicative of text difficulty. At TTIC, Trang is working with Karen Livescu, Kevin Gimpel, and Mohit Bansal on incorporating speech information to improve constituent parsing. Trang is broadly interested in research topics involving both speech and NLP, with a preference for applications in language disorders and social science. She is from Hanoi, Vietnam, and likes to run in her free time.

61

FINANCIAL REPORTS Independent auditing agency: Plante Moran | 225 W. Washington St. Suite 2700 | Chicago, IL 60606

62

63

FACULTY RESEARCH HIGHLIGHTS

64

Finding the invisible, and improving accuracy, in genetic transcription The latest RNA-sequencing techniques create an avalanche of fragmented genetic data. TTIC’s Stefan Canzar leads a team using advanced algorithms to reassemble that information and detect missing patterns. Accurately reconstructing the basic protein patterns in the genetic blueprints of life can help researchers understand cellular biology and identify specific patterns in genes that produce diseases. The preferred method of converting samples of cellular RNA into genetic data is RNA sequencing (RNA-seq), which creates hundreds of millions of small data fragments, known as reads. Piecing together the resulting mass of highly fragmented data into the full-length molecular sequences, known as transcripts, is difficult. But a group of scientists from Toyota Technological Institute of Chicago (TTIC), Freie Universität Berlin, and Centrum Wiskunde & Informatica, have developed a way to use advanced algorithms to reconstruct RNA sequences more accurately and comprehensively than other current methods. RNA stands for ribonucleic acid, a complementary molecule to DNA that’s found in all living cells. While every cell of the same organism contains the same DNA molecules, their genes are transcribed into different RNA sequences depending on cell type. In RNA-seq, scientists break the content of a cell into tens of millions to hundreds of millions of short sequences of 100 to 150 bases, called reads. Reads can be pieced together to reconstruct an RNA transcript. But piecing the reads back together in correct combinations is difficult, because they often contain little information about where they belong. “It is a big jigsaw puzzle,” says Stefan Canzar, a research assistant professor at TTIC. “You want to figure out the big picture—these RNAs—but you only have these small fragments, these puzzle pieces.” Further complicating matters, one subgroup of transcripts is particularly hard to reconstruct. When transcripts are rare in the sample, key data about the connectivity of basic informational blocks that make up transcripts, known as exons, may not be evident

from RNA sequencing. The researchers call these “invisible transcripts.” Existing methods being used to reassemble RNA ignore such transcripts for computational reasons. The technique pioneered by Canzar and his colleagues, Comprehensive Isoform Discovery and Abundance Estimation, or CIDANE, mixes techniques from machine learning and combinatorial optimization to reconstruct transcripts. CIDANE can also use existing information from known, and experimentally validated, gene structures to improve the accuracy of the RNA assembly. While such information is not necessary, CIDANE can use any existing data to improve its ability to reconstruct RNA. “Any prior knowledge can help in piecing together the genetic puzzle,” says Canzar. CIDANE is also able to reassemble the elusive invisible transcripts, detecting patterns that other techniques miss, Canzar says. Using a technique from large-scale optimization, CIDANE can discover those transcripts using an optional stage of the algorithm that kicks in on-demand when invisible transcripts appear to be involved. While the importance of invisible transcripts is still uncertain, by recognizing them, CIDANE gives genetics researchers more and better information about their existence and value. Ultimately, CIDANE is more accurate than all existing reconstruction methods, Canzar says. The most sensitive reassembly technique after CIDANE, when applied to human blood and monocyte samples, correctly pieced together 11,473 and 11,117 transcripts respectively. CIDANE correctly predicted 14,885 and 14,254, which is 80 to 90 percent more transcripts than the third most sensitive, and most widely used, reassembly technique. Stefan Canzar, Sandro Andreotti, David Weese, Knut Reinert and Gunnar W. Klau, “CIDANE: comprehensive isoform discovery and abundance estimation,” Genome Biology, January 2016.

65

Hunting for hidden structures in complex networks TTIC’s Julia Chuzhoy studies complicated graphs, aiming to create faster and more accurate algorithms. Finding the best driving route, connecting with friends on Facebook, buying a cheap airline ticketall of these rely on what computer scientists call “graphs” or networks, the mathematical term for a sets of nodes connected by edges. Because graphs are used so widely, better understanding their structures leads to superior algorithms, and to better and faster computer programs. TTIC’s Julia Chuzhoy is trying to deepen our understanding of graphs by looking for hidden structures lurking in complex networks. Imagine routing traffic across a graph, which could represent the streets of a city or a corporate computer network. “We would like to find pathways for traffic to follow, while keeping the load on the graph edges as low as possible,” says Chuzhoy. Take a city such as Phoenix, which typically closes some roads in the spring for construction and sets up detours. How should the city reroute traffic so that vehicles continue to efficiently move through the network? In graphs such as the one formed by Phoenix’s road system, there is a hidden structure: a grid. In Phoenix, a good number of roads run either North to South or East to West. The street intersections are the grid nodes, and the road segments connecting the intersections are its edges. Grid graphs are among the simplest to analyze, but they are still rich in interconnections, so routing traffic on them is easy and convenient. The larger the grid, the more traffic it can carry. Mathematicians Neil Robertson of the Ohio State University and Paul D. Seymour of Princeton famously showed in 1986 that any graph that’s complex enough contains a grid. But the grids they found were too tiny to be useful when routing large amounts of traffic across a network. Since then, the question of whether complex graphs contain large grids has remained a central open question in graph theory.

66

Graphs that are easy to analyze yet rich enough to offer many routing options are called crossbars. In her recent work, Chuzhoy showed for the first time that if a graph is complex enough, it must contain a certain kind of a crossbar, which she calls a “tree-ofsets system.” This crossbar is so big that it can be used to route large amounts of traffic across networks with vastly better load balancing than before. This represented a huge step forward in understanding network routing, but it left open the bigger question of whether another kind of crossbar—a large grid—exists in complex graphs. “While any crossbar would work for routing problems, grids have many more known uses, so they are in some sense ‘universal’ or ‘all-purpose’ crossbars,” Chuzhoy says. In a follow-up work with Chandra Chekuri, a colleague from University of Illinois, Chuzhoy shows that the tree-of-sets system helps not only with routing, but also with finding large grids, answering a question that had been open for almost 30 years. The grids that Chuzhoy and Chekuri find are quite close in size to the best one could hope for. This result sheds new light on the structure of complex graphs and creates far more powerful tools for using them. For example, it’s already led to better and much faster algorithms for many graph-based problems. While theoretical, Chuzhoy’s research could ultimately be used to help design algorithms that make communication and transportation networks more reliable and solve an array of other practical problems—from finding more efficient designs for electronic circuitry, for instance, to planning complex maneuvers of swarming robots. Julia Chuzhoy, “Routing in Undirected Graphs with Constant Congestion,” SIAM Journal on Computing, 2012. Chandra Chekuri and Julia Chuzhoy, “Polynomial Bounds for the Grid-Minor Theorem,” Symposium on Theory of Computation, September 2014.

Can computers learn to color? A machine-learning system developed by TTIC researchers can accurately colorize black-andwhite images. Human imagination is adept at instilling color into a black-and-white photograph. Our familiarity with the visual world is so ingrained that a child completing a coloring book understands what makes a realistic palette. Sky is usually blue, but other objects, such as a car, could be any one of many colors. In pursuit of building machines that can interpret the visual world as well as humans, some researchers see color as a starting point. Gustav Larsson, a PhD student in computer science at the University of Chicago, and TTIC's Michael Maire and Gregory Shakhnarovich, have found a way to teach machines to accurately predict colors using new machine-learning techniques based on artificial neural networks. This new method could be used to automatically colorize historical images and films, and create computer graphics tools for artists. But ultimately, the work could lead to a method of priming a robot's artificial visual system so it could recognize and navigate its environment. “Automatic colorization also serves as a proxy measure for visual understanding,” the researchers write. Rather than start from scratch, Larsson, Maire, and Shakhnarovich began with an existing neural network that could recognize 1,000 different types of objects. They then focused on modifying this network to predict a color for every pixel in a grayscale image - even if the pixel was part of an unrecognized object. To do so, they trained the network using a database of 1.2 million color photos. Converting each to grayscale enabled them to present the network with example pairs of a grayscale input image and corresponding desired color output image. Teaching the network proceeded via a routine of learning from mistakes. The researchers used a network with a deep architecture: one layer of neurons is connected to the input, while a second layer of neurons is connected to the first layer. This pattern repeats for many layers, the last of which is treated as the output. The strength of each connection is a parameter that can be adjusted. The network contained a total of almost 150 million such parameters, and as a whole those parameters determined the network's behavior. Learning from each example involved tweaking the parameters so that the network's output became more similar to the actual color image. This required propagating parameter adjustments backwards through the network, from last to first layer. After taking a week to iterate through all 1.2 million examples 10 times, the network was well tuned to the task of colorization. The researchers point to two custom aspects of their architecture design that were key to accurate colorization. First, they added shortcut connections from intermediate layers of the network to deeper layers. For vision tasks, neurons in early layers typically learn to respond to local image structures, such as edges and corners. As one moves to subsequent layers, neurons respond to a larger context and increasingly abstract concepts: textures, object parts, objects, and scenes. When deciding on a color, the shortcut connections “allowed the network to combine the ‘what’ and ‘where’ aspects of image understanding,” says Shakhnarovich. Second, the network’s predictions appeared in the form of color histograms, rather than as exact color values. This allowed the system to hedge its bets, especially for objects such as clothes or flowers that typically occur in different colors. As a side benefit, the system’s histogram output made it possible for it to automatically generate several different plausible colorizations - a potential boon to artists wanting to manipulate their work with different color palettes.

67

Can computers learn to color?, continued.

According to quantitative measures of color accuracy, the researchers’ method works significantly better than those of its predecessors. Many earlier attempts also required users to advise the computer, either by scribbling some suggested colors on the image and having the system fill in the rest, or by presenting related photos from which to transfer colors onto the target image. Outperforming these approaches, while being fully automatic, means the method vastly reduces the human labor required to achieve accurate colorization. With a working system in place, the researchers went back to the start in order to expand their goals: could they remove the requirement to first know about 1,000 objects? To pick colors, the network must have identified many uncategorized objects - the set of 1,000 objects only provided a head start. If the colorization task alone could teach the network about the visual world, their system would succeed without this head start. To test this hypothesis, the researchers repeated their training procedure starting from a network with connection strengths that were randomly chosen, rather than borrowed from a network that already knew about objects. The resulting trained network was able to colorize images just as well as the original network, proving that it wasn’t necessary to start with object knowledge. The next test was to probe what the network actually knew about the visual world. “We want to see if learning to correctly predict the color makes it easier to learn about different types of objects in images,” Shakhnarovich says. They gave the colorization-trained network the additional task of labeling pixels according to object category (for example, car, person, chair, or table). The colorization network proved capable of labeling objects, performing better at this than a baseline network that had not been first taught to colorize. This result suggests colorization is a gateway to building other systems that understand images. And for colorization, the learning process itself can be automated with an endless supply of color images on the Internet. The researchers hope that future machine vision systems will learn on their own, like a child playfully exploring without any instruction from parents. Gustav Larsson, Michael Maire, and Gregory Shakhnarovich, “Learning Representations for Automatic Colorization,” Proceedings of the European Conference on Computer Vision, preprint, arXiv:1603.06668, 2016.

68

69

GOVERNANCE

Board of Trustees

Robert Barnett Partner, Williams & Connolly LLP Ranked Number One, Washingtonian Magazine’s list of “Washington’s Best Lawyers.” Executive Committee Member, Williams & Connelly LLP Senior Counsel, Board of Trustees of the John F. Kennedy Center for the Performing Arts. (President-appointed member.) Trustee since April 2006 Rita Colwell Chairman, Canon US Life Sciences, Inc. Distinguished Professor, University of Maryland College Park and Johns Hopkins University 11th Director of the National Science Foundation, 1998-2004 Authored or co-authored 16 books and more than 700 scientific publications A geological site in Antarctica, Colwell Massif, named in recognition of her work in the Polar Regions Trustee since September 2008 Sharon Darling President and Founder, National Center for Family Literacy Frequent keynote speaker: Businessweek Fortune 500 Forum and the National Governors Association Recipient of the 2002 National Humanities Medal awarded by President and Mrs. George W. Bush, and the Albert Schweitzer Prize for Humanitarianism from Johns Hopkins University Serves on the boards of: the Barbara Bush Foundation for Family Literacy, the National Fund for Excellence in American Indian Education, Corporation for Public Broadcasting’s Ready to Learn, and the Heart of America Foundation Trustee since April 2007 Robert A. Fefferman Max Mason Distinguished Service Professor, Division of the Physical Sciences, University of Chicago Former Dean, Division of the Physical Sciences, University of Chicago Former Chairman, Department of Mathematics, University of Chicago Recipient, Quantrell Award for Excellence in Undergraduate Teaching, and University of Chicago Sloan Foundation Fellow Trustee since October 2003 Sadaoki Furui President, Toyota Technological Institute at Chicago Professor Emeritus, Tokyo Institute of Technology Professor, Academy for Global Leadership, Tokyo Institute of Technology Former Director of University Library, Tokyo Institute of Technology Former Dean of Graduate School of Information Science and Engineering, Tokyo Institute of Technology Former Director of Furui Research Laboratory, NTT Human Interface Laboratories, Japan Former Director of Speech and Acoustics Laboratory, NTT Human Interface Laboratories, Japan Trustee since April 2013 Eric Grimson Chancellor for Academic Advancement, Massachusetts Institute of Technology Bernard Gordon Chair of Medical Engineering at MIT Lecturer on Radiology at Harvard Medical School and at Brigham and Women's Hospital Former Education Officer for the Dept. of Electrical Engineering and Computer Science at MIT; Associate Department Head; Head of the Depart. of Electrical Engineering and Computer Science. Trustee since July 2015

70

Alexis Herman Chair and Chief Executive Officer, New Ventures, LLC Appointed by President Jimmy Carter, became the youngest director of the Women's Bureau in the history of the Labor Department US 23rd Secretary of Labor and first African American to lead the US Department of Labor Former member of the National Economic Council Serves on the boards of: Cummins Inc., Entergy Inc., MGM Mirage, Coca-Cola Company Former chairwoman of the Coca-Cola Company’s Human Resources Task Force Board member of the Clinton Bush Haiti Fund Trustee since October 2012 Masanori Kashiwara Executive Advisor, Toyota Technological Institute Member of the Board of Directors, Toyota School Foundation Former Chief Administrative Officer, Toyota Technological Institute Former Vice President, Toyota Motor North America, Inc. Former Secretary and Treasurer, Toyota Motor Corporate Services of North America, Inc. Trustee since October 2009 Edward Kolb Dean, Division of Physical Sciences, University of Chicago Arthur Holly Compton Distinguished Service Professor Member, Enrico Fermi Institute Board Member, Giant Magellan Telescope, 2010-present; Adler Planetarium, 2010-present Trustee since October 2013 Jim Merz Frank M. Freimann Professor Emeritus of Engineering, Concurrent Professor of Physics, University of Notre Dame Fellow, American Physical Society Trustee since July 2015 Nelson Morgan Professor-in-residence (emeritus) Electrical Engineering and Computer Science Dept., University of California, Berkeley Emeritus Director, International Computer Science Institute Former Editor-in-chief of Speech Communication Fellow of the IEEE and of the International Speech Communication Association (ISCA) Trustee since April 2015 Hiroyuki Sakaki President, Toyota Technological Institute Professor Emeritus in 2007, Institute of Industrial Science, University of Tokyo Former Vice President of Toyota Technological Institute (Nagoya, Japan) in 2007 and promoted to President in 2010 Awarded the National Recognition as a Person of Cultural Merit, Japan Academy Award, Leo Esaki Award, Heinrich Welker Award, Medal of Purple Ribbon from the Emperor of Japan, IEEE David Sarnoff Award, Fujiwara Prize, Japan IBM Science Award, and the Hattori-Hoko Award Trustee since October 2010 Richard J. Samuels Ford International Professor, Dept. of Political Science, Massachusetts Institute of Technology Director, MIT Center for International Studies Founding Director, MIT Japan Program Former head of the MIT Political Science Department, Vice-Chair of the Committee on Japan of the National Research Council, and chair of the Japan-US Friendship Commission. Trustee since July 2015

71

Toshiaki Taguchi Advisor, Toyota Motor Corporation Former President and CEO, Toyota Motor North America, Inc. Former Executive Vice President, Toyota Motor Corporation Former Board of Directors of Japan Society, the Japanese Chamber of Commerce and Industry of New York and the Nippon Club Trustee since October 2002

Masatami Takimoto Chairman of the Board of Trustees, Toyota Technological Institute at Chicago Chairman of the Board of Directors & the Board of Trustees, Toyota School Foundation Special Advisor, Toyota Central R&D Labs., INC. Former Executive Vice President, Toyota Motor Corporation Trustee since October 2011

Tatsuro Toyoda Chairman Emeritus

Mitsuru Nagasawa President Emeritus

Mark Hogan (Advisor to the Board) Director, Toyota Motor Corporation

Kosuke Ikebuchi (Advisor to the Board) Advisor, Senior Technical Executive, Toyota Motor Corporation

72

LEADERSHIP Dr. Sadaoki Furui, President Mr. Masashi Hisamoto, Treasurer and Secretary to the Board Ms. Jessica Johnston, Chief Financial Officer Dr. David McAllester, Chief Academic Officer Ms. Chrissy M. Novak, Secretary of the Institute

ADMINISTRATION Adam Bohlander, Director of Information Technology Erica Cocom, Student Services Assistant Mary Marre, Administrative Assistant Amy Minnick, Human Resources Generalist, Immigration Specialist Chrissy M. Novak, Administrative Director of Graduate Studies, Publications Anna Ruffolo, Director of Operations Chendi Wu, Senior Accountant

Equal Opportunity Statement TTIC, in admissions, employment and access to programs, considers all faculty, staff and students on the basis of individual merit and without regard to race, color, religion, sex, sexual orientation, national or ethnic origin, age, disability, or any other legally protected status.

73

SPECIAL THANKS The External Advisory Committee Eric Grimson, Chancellor and Professor of Computer Science and Engineering, Massachusetts Institute of Technology Takeo Kanade, UA and Helen Whitaker University Professor, Robotics Institute, Carnegie Mellon University Richard Karp, Professor of Electrical Engineering and Computer Science, University of California, Berkeley Éva Tardos, Jacob Gould Schurman Professor of Computer Science The University of Chicago greater community Booth School of Business Chicago Center for Teaching Computation Institute Department of Computer Science Department of Mathematics Department of Statistics Faculty and Administration of the Division of Physical Sciences Office of International Affairs Office of the Bursar Physical Science Division- Local Business Center PSD Graphic Arts Registrar’s Office Staff of the Regenstein and Eckhart Libraries Student Health and Counseling Services University Research Administration University IT Services The professionals at the 6045 S. Kenwood Avenue building Dr. Sunil Ahuja of the Higher Learning Commission The Toyota Central R&D Labs, Inc. Toyota Motor Corporation The Toyota Technological Institute (Nagoya, Japan)

74

75

6045 S. Kenwood Ave. Chicago, IL 60637

www.ƫc.edu

@TTIC_Connect