Essence: A Machine Learning Approach to Scalable and Energy ...

21 downloads 0 Views 14MB Size Report
Compressive Sensing. • Scalability and Energy Efficiency ... Power of the IoT+CrowdSensing ... On phone, on broker (SenseDroid, SATWARE). • Techniques ...
Essence:  A  Machine  Learning  Approach         to  Scalable  and  Energy-­‐Efficient                       Sense-­‐making  for  Internet-­‐of-­‐Things  (IoT)  

Santanu  Sarma   University  of  California  Irvine   September  2015    

Overview   •  •  •  •  •  • 

Sense-­‐making  in  IoT       Problem  Descrip9on     Compressive  Sensing     Scalability  and  Energy  Efficiency   A  Middleware  Perspec9ve     Summary  

IoT  Symposium,  Oct  2015  

 

©  S.Sarma  

2  

Internet-­‐of-­‐Things  

IoT  Symposium,  Oct  2015  

©  S.Sarma  

3  

Internet-­‐of-­‐Things  

[Vermesan2011]  

IoT  Symposium,  Oct  2015  

©  S.Sarma  

4  

IoT  Trends  

IoT  Symposium,  Oct  2015  

©  S.Sarma  

5  

What  is  Sense-­‐Making?   •  An  informa9on-­‐theore9c  perspec9ve  to  mean   the  process  of  understanding  the  connec1ons   (among  people,  places,  environment,  objects,   and  events)     –  in  order  to  an9cipate  their  trajectories     –  develop  meaningful  insights     –  act  effec9vely  using  empirical  data  /insights    

IoT  Symposium,  Oct  2015  

©  S.Sarma  

6  

IoT+CrowdSensing

Pushing  toward  more  interven9on  

7  

IoT  Symposium,  Oct  2015  

©  S.Sarma  

Power  of  the  IoT+CrowdSensing   •  Using  mobile  crowdsensing  to   –  Leverage  already  deployed   smartphones     –  Extend  the  ranges  of  exis9ng  in-­‐situ   sensors   –  Send  mobile  users  to  specific   loca9ons  

•  Crowdsensing  broad  use  cases   –  Disaster  and  emergency  response   –  Personal  health  monitoring  and   wellness   –  Smart  spaces  and  their  effec9ve   u9liza9on   IoT  Symposium,  Oct  2015  

©  S.Sarma  

[Yuen2011]   8  

Sensing  -­‐>  Sense-­‐making   Severity   Alert  System  

Personal  Sensing  to  indicate  Fall  detec9ons,   injury  severity,  alerts  in  old  age  people  to   provide  scalable  health  care     IoT  Symposium,  Oct  2015  

©  S.Sarma  

9  

Sensing  -­‐>  Sense-­‐making   Radia9on  field  near  Fukushima  

Hazardous  gas  in  campus  

Spa9al  Field  Sensing  With  Mobile  Sensors  

IoT  Symposium,   ct  2015    Latest  Informa9on   Crisis   Map  SOhowing  

©  S.Sarma  

10  

Sensing  -­‐>  Sense-­‐making •  Avoiding  congested  streets  in  a  city   •  Finding  the  most  popular  booth  in  a  fair   •  Searching  for  the  ride  with  shortest  lineup  in  an   amusement  park  

IoT  Symposium,  Oct  2015  

©  S.Sarma  

11  

Sense-­‐Making  :  Purpose  &  Goals   u A  Framework  for  Sense-­‐making  using  mobile   phone  &  infrastructure  sensors  to  derive  insights   u Powerful  addi9onal  sensing  abili9es  and   features  for  community  of  users  by  community   of  users     u Understand  user  and  group  context  efficiently   u Develop  a  efficient  mechanism  (for  energy  ,   performance)     IoT  Symposium,  Oct  2015  

©  S.Sarma  

12  

The  Problem  –  A  cross  layer,  end  to  end   issue   §  Several  barriers  and  huge  investment  of  9me   to  build  sense-­‐making  applica9ons     §  Lack  of  a  framework  hinders  ease  and  speed   the  development  of  sense-­‐making  apps   §  Non-­‐Scalable,  Ad-­‐hoc,  non-­‐standardized  API     §  Unsupported  network  infrastructure,  and   configura9ons    

IoT  Symposium,  Oct  2015  

©  S.Sarma  

13  

Middleware  Plalorms  and  Techniques      for  Sense-­‐making     •  On  phone,  on  broker    (SenseDroid,  SATWARE)   •  Techniques  implemented  in  middleware   –  Compressive  and  Collabora9ve  Sensing     –  Virtual  Sensing  for  Sense-­‐making   –  Seman9cs  Driven  Sensing  and  Actua9on    

•  Combining  In-­‐situ  Sensors  with  Mobile   Crowdsensing  

IoT  Symposium,  Oct  2015  

©  S.Sarma  

14  

Essence  Hierarchical  Architecture  

IoT  Symposium,  Oct  2015  

©  S.Sarma  

15  

Essence  Distributed  Middleware     APPS$2$ APPS$N$

APPS$1$ Cloud Mobile$Node$

S1$

Sn$

Sensing$&$ Sampling$

Manager$

Context$ Processing$ &$Fusion$$

Privacy$&$ se>ngs$

Communica.on$

Query$+$ Storage$

Query$ Query$&$$ Response$

Analysis$&$Processing$

AP

Communica.on$

Data$Collec.on&$ Comp.$Sampling$

Sn$

Sensing$&$ Sampling$

Manager$

Context$ Processing$ &$Fusion$

Privacy$&$ se>ngs$

Communica.on$

Query$+$ Storage$

Collabora.on$

Query$&$$ Response$

S1$

Response$

Broker$

Infrastructure$Sensing$$

…….$

Query$+$ Storage$

IoT  Symposium,  Oct  Mobile$Node$ 2015  

Manager$

©  S.Sarma  

S1$

S2$

…….$

Infrastructure$Sensors$

Sm$

16  

Sense-­‐making  Using  Compressed  Sensing   •  A  random  sampling  technique  that  can  represent  Sparse  signal  with  few   random  measurements   •  Represents  a  Sparse  Signal  with  few  salient  coefficients  in  a  transformed   domain   •  Integrates  sensing,  compression,  processing  based  on  new  uncertainty   principles  

IoT  Symposium,  Oct  2015  

©  S.Sarma  

17  

Compressive  Sensing  Comparison  

[Baraniuk2008]   IoT  Symposium,  Oct  2015  

©  S.Sarma  

18  

Reconstruction##Error#(MSE)#

Collabora9ve  Compressive  Sensing  

Legend

Sink Node(Broker)

Sampled Mobile Sensor

Mobile Node

No#of#Measurements##

Traded-­‐off   Number  of  Measurement                                  Accuracy  of  Sensemaking       Number  of  Measurement                                    Energy  Consumed  in  Sensing   Accuracy  of  Sensemaking                                    Scalability  and  Coverage     IoT  Symposium,  Oct  2015  

©  S.Sarma  

19  

Execu9on  Time  for  Sense-­‐Making  

IoT  Symposium,  Oct  2015  

©  S.Sarma  

20  

Energy  Consumed  in  Sense-­‐Making  

IoT  Symposium,  Oct  2015  

©  S.Sarma  

21  

Applica9on:  Sparse  Temperature  based   West  Nile  Virus  (WNV)  Predic9on        

[Kaggle.com]  

IoT  Symposium,  Oct  2015  

©  S.Sarma  

22  

Research  Direc9ons   •  Energy  Efficiency   –  Exploit  collabora9ve  &  compressive  sensing  for   energy  efficiency  

•  Incen9ve  Mechanisms   –  Device  incep9ves  for  par9cipa9on  and  collabora9on  

•  Privacy  Regula9on   –  Facilitate  privacy  preserving  incen9ves  

•  Heterogeneity  in  Mobile  Cloud   –  Use  and  exploit  heterogeneity  of    sensors  and  devices     IoT  Symposium,  Oct  2015  

©  S.Sarma  

23  

Summary   •  Studies  a  new  class  of  IoT  based  Sense-­‐making  using   machine  learning  

–  Geospa9al  informa9on  gathering   –  using  combina9on  of  crowdsourcing  and  infrastructure  sensing    

•  Proposes  compressive  sensing  based  Sense-­‐making     •  Simula9on  results  are  encouraging   •  Poten9al  Extensions  

–  Implemen9ng  a  working  prototype   –  Guide  the  workers  to  shoot  photos  using  augmented  reality   –  Quality  assurance  and  cheat  detec9on  mechanisms  

•  Designed  for  collec9ng  spa9al-­‐temporal  informa9on,  but   can  be  extended  for  event  detec9on     IoT  Symposium,  Oct  2015  

©  S.Sarma  

24  

THANK  YOU  

IoT  Symposium,  Oct  2015  

©  S.Sarma  

25  

EXTRA  SLIDES  

IoT  Symposium,  Oct  2015  

©  S.Sarma  

26  

Sensors  In  Mobile  Phones  

•  MEMS  &  sensors  for  cell  phones,  expanding  from  $  3.5  bn  in  2009  to  $7.9  bn   in  2015  [Yole  Developpement]   •  Smartphone  sensors  to  be  $  6  bn  business  by  2016  [Juniper  Research]   •  44  %  of  the  mobile  phones  will  be  smartphones  in  2015   •  7x  increase  in  mobile  health  apps  from  2010  to  2011   •  moaon  sensor  in  smartphones  and  tablets  will  expand  to  $  US  2.1  billion  in   IoT  Symposium,   Oct   2015   .Sarma  billion  in  2011  (IHS  iSuppli)   2015  w ith   a  25.3  %  CAGR,  up  from  ©  $S1.19  

27  

Mobile  Phone  Trends   •  Mobile  subscripaon  5.96  billion   2011  esamate   •  Smartphones  (487.7  million)   exceeding  PCs  (414.6  million)   •  More  Mobile  Internet  Users   Than  Wireline  Users  in  the  U.S.   by  2015   •  Smartphone  and  bandwidth  cost   reduces   •  Smart  devices  contribute  to   more  than  90%  of  mobile  data   traffic   IoT  Symposium,  Oct  2015  

©  S.Sarma  

28  

Mobile  Sensors  Trends  

Source:   EMS  Market  Tracker,  April  2014.     IoT  Symposium,   Oct  2015  IHS  Consumer  &  Mobile   ©  M S.Sarma  

29  

Mobile  Data  Delivery  Everywhere   The  exploding  number  of  apps  is   driven  by  a  huge  upack  in  the   number  of  smart  devices  

Smart  devices  contribute  to  more   than  90%  of  mobile  data  traffic  

~55%  

IoT  Symposium,  Oct  2015  

Cisco’s  report  2014  

©  S.Sarma  

30  

IoT  for  Emergency  Use  Cases     • Earthquakes • Hurricanes • Tornadoes • Energy/utility outages • Fire hazards • Hazardous materials releases • Terrorism/

IoT  Symposium,  Oct  2015  

©  S.Sarma  

31  

IoT  for  Emergency  Response  

During  Fire  accidents  can  cause  electric  power  failure.  Mobile  broadcast  can  be   used  to  provide  direcaons  to  the  users  about  rescue  operaaons.   IoT  Symposium,  Oct  2015  

©  S.Sarma  

32  

IoT  for  Emergency  Response  

Emergency  situaaon  Automaac  Altering  can  be  used  to  inform  family,  rescue   teams,  or  nearby  cars  /  passengers  in  case  of  accidents.   IoT  Symposium,  Oct  2015  

©  S.Sarma  

33  

Suggest Documents