2018 BINTI AI Trading Service Network
Page | 0
Table of Contents SITLAB Vision and Overview .................................................................................................................... 2 SITLAB Cloud based AI Trading Service Network (BINTI) ........................................................................ 3 Architecture Overview ............................................................................................................................ 3 Distribution of Intelligence .................................................................................................................. 3 Infrastructure Architecture and Deployment ..................................................................................... 4 Artificial Intelligence Elements (AI Services) ....................................................................................... 6 Trading Elements ................................................................................................................................. 8 Private Cloud Infrastructure Elements .............................................................................................. 10 Ultra-Low Latency Competitive Artificial Intelligence (AI) and/or Machine Learning ...................... 12 Development and Operational Implementation ................................................................................... 14 BINTI deliverable goals ...................................................................................................................... 14 BINTI Strategic Technology Partners ................................................................................................. 15 BINTI Roll out strategy....................................................................................................................... 16 Cloud based AI Trading Service Network (BINTI) Summary .................................................................. 17
Page | 1
SITLAB Vision and Overview The digitalization in the financial industry boosts the data volume, velocity, variety, and veracity. Today’s retrospective data analysis and machine learning systems, let alone human experts, cannot analyze this growing data flood fast enough to understand it in the actual context and extract information for decision making. Also, more and more data are of real-time nature, which is inevitably lost if it is not constantly processed. Therefore, information mining and informed decision making is one of the major challenges of the raising new digital age in the financial industry and is particularly relevant for financial trading and asset management. Our Vision: Is to not only to collect trading data and process it and analyze it (in real time) locally to where those trading decision need to be made, but to also distribute that “intelligence” gathered across a wider neural network of trading venues so that localized trading can be influenced by (and learn from) the wider and more global knowledge of trading decisions on other markets and also regional and worldwide geo-political data. For this purpose, we designed and have started to develop BINTI. BINTI is a private/hybrid cloud based intelligent trading service network that is operated and managed by SITLAB that has specialized and distributed trading hardware, software and services collocated at major trading venues (globally) to perform “intelligent” trading. In addition, there is a distributed set of regional sites where “intelligent data” is being collected from multiple (regional) local trading venues and regional digital geo-political data that can impact trading decisions. Consider this a hierarchy of knowledge gathering and actions across a neural network based on trading activities at the periphery (edge) of the network through regional data centers to the core of the network at a designated worldwide datacenter. The BINTI service network will be multi-client with SITLAB selling a variety of services for customers trading via this network. Customers of the service network (Traders and/or Brokers) will benefit from improvement of sell- & buy-side client bottom line combined with 3rd party technological integration, optimization & automation which allows them to achieve an overall client revenue improvement in addition to an overall cost reduction in technology Capex, Opex and resource spending. The implementation of the BINTI Service network will be done in multiple phases over a three year period. SITLAB’s key intellectual property will be the ability to develop a series “Trading agents” executing at various locations and levels of “intelligence” hierarchy within their optimized private cloud-based service network. They will sell and build out of their customer’s own private network within that service network and any additional customized Trading agents that their customers may want for their environment.
Page | 2
SITLAB Cloud based AI Trading Service Network (BINTI) Architecture Overview Distribution of Intelligence The BINTI trading system utilizes AI technologies to provide Machine Learning Trading recommendations:
These recommendations are Geographically and Hierarchically compounded and distributed to the to the local BINTI trading machines affected. The trading recommendations are based on localized AI Database monitoring and processing by AI engines to forward on those recommendations.
The Hierarchy is based on a geographically distributed set of BINTI AI centers broken in tiers:
A BINTI (Colocation) AI center that is situated at each trading exchange colocation center A number of BINTI (Regional) AI center(s) that are situated regionally (NA, EMEA, APAC) that services all the exchange colocations within that region. A BINTI AI center that is situated at a WW level that services all the BINTI Regional AI centers.
The compounding and distribution of the intelligence allows a much more targeted set of Alpha triggers on what the individual trader at the specific exchange and markets are trading on will/can be impacted!! It should be noted that the intelligence is used on a Trader by Trader basis based on the venue(s) they will be trading at and based on the breadth of “intelligence” they will be using locally at both regional and WW AI Centers.
Figure 1. BINTI Distributed Intelligence
Page | 3
Infrastructure Architecture and Deployment The BINTI trading system utilizes a private cloud infrastructure for each trader/customer. That infrastructure expands from the Trading Customer’s own premises where the trading intelligence information will be made visible and that new trading strategies can be implemented and executed from. SITLAB has its own private cloud infrastructure into BINTI based out of KTH’s facilities in Stockholm, Sweden. This private based architecture will be leveraged to deploy new trading agents across the BINTI infrastructure at Colocation, Regional and WW levels, as well as providing upgrades to customer trading agents being developed by SITLAB. The implementation and deployment of this private cloud infrastructure will be the responsibility of a key technology partner under contract with SITLAB. That partner will also have the responsibility to architect, deploy, manage and maintain the BINTI Service Network under contract with SITLAB for a minimal of the three year period.
Figure 2: SITLAB AI Trading Service Network
The BINTI Service Network is made up three major logical parts:
The Artificial Intelligence (AI) elements: o Runtime environment that provides Trading agents that learn from the local environment to recommend optimal trading strategies based on that information. o The runtime data that provides the basis for learning is distributed across multiple geographical locations and has access to a variety of geo-political data. o Development environment that is based at SITLAB in Stockholm and is used for basic testing and development of new trading agents.
Page | 4
The Algorithmic Trading elements this is made up of various blocks: o Distributed Time Infrastructure (required for synchronization and timestamping) to meet regulatory requirements. o Core Network infrastructure for management and distribution of Market Data and Intelligent Order execution. o Capture and Analytics Block that is required to capture raw information and provides real time analytics associated with the Trader’s order execution and market data feeds. This is required for 3rd party audit reasons and also useful for real time analytics and post processing analytics. This also feeds analytics data into a trading database for historical back testing and for ingestion by the machine learning trading agents. o Trading Database this is the storage medium for the streaming of real time analytical data (trade executions and market data feeds). It is also the repository for the Machine Learning Trading agents and the Trading execution infrastructure for ongoing trading logs and back data testing etc. o Trading Execution Infrastructure that executes various trading strategies based on the real time data coming in. The trading strategies are tuned over time based on the AI trading agents feedback and when the customer wants to put them into production.
The Private Cloud Infrastructure: o The complete Runtime AI and Trading infrastructure operates and executes under this infrastructure. o From the trader’s (customer) perspective its distributed across multiple co-location venues to multiple trader’s venues at a global level. o From SITLAB perspective it provides the private cloud to Distribute Machine Trading agents that have been developed at the SITLAB facility to the various Colocation, Regional and WW sites. It also provides the automation for moving “Intelligent” KPIs across the distributed infrastructure delivering data where its needed to be analyzed in the “Intelligence Hierarchy”. o The infrastructure provides enterprise quality resiliency, security and performance optimization. It is based on an Intent driven network environment. o The infrastructure provides scalable upgrades to increase processor, memory, networking and storage capability. o It also incorporates the use of specialized appliances (Capture and Analytics and AI Appliances) for the automated deployment and management of the said appliances as part of the overall infrastructure.
Page | 5
Artificial Intelligence Elements (AI Services) The AI framework is based on an agent framework, where agents are learning autonomous decisions by interacting with an uncertain environment in a trial an error manner. The learning is initiated with an offline training phase to collect special knowledge about the specific trading activity the agent should perform, followed by constant online learning to follow the market more closely. The AI models describe •
How the agents see the market, which also includes how often they inspect the market changes
•
How agents are rewarded for successful decisions
•
How they learn offline, i.e. from historical data
•
How they learn online, i.e. if new data coming in
The AI models use recent innovations from Deep Learning developed at various labs, including Google Deep Mind and OpenAI, as well as some proprietary innovations tailored to multivariate time series and the financial trading domain, and techniques to scale up the agent training process. The Runtime Platform: The runtime platform is equally important. It takes care of the data handling, the agent training, the agent deployment and the scalability.
Exchange
Apache NiFi dataflow management
Bank
User Interface (UI)
Data Provider
Data Ingestion
Agent Manager
Apache Kafka Storage HDFS AWS S3
Stream processing: Live stats Feature extraction LOB reconstruction
Agent bootstrap training
Trading system
Data live UI
Figure 3: AI Runtime Platform Infrastructure Page | 6
First the platform is designed to be cloud native. The backbone of the platform is Apache Kafka, which handles the data streaming to the agents. It also handles online data collection and offline data loading from HDFS storage. The stream processing component handles various data transformation and data preparation. The infrastructure is designed to be scalable and extensible. Apache Avro data schemas are used to describe the data and allow efficient storage of the data in binary form without the need to store meta data.
Sitlab Agent Designs
Sitlab Trading Execution Backend
AI and Deep Learning based Models
Technology Platform
Reactive end-to-end autonomous trading agents
Figure 4: AI Powered Trading Agent AI Powered Trading Solution:
Training o
Agents trained with Deep Reinforcement Learning
o
Complementary trading algorithms based on signals
o
Training infrastructure and models implemented with DL framework
o
GPU / FPGA infrastructure tuned for training and for fast inference of deep agent models
o
Infrastructure used also in online training of pre-trained agents
o
Use generative models to extend training data
Trading o
Provide order execution infrastructure and link to exchange and streaming market data
o
Agents exchange market and position information with algorithmic trading automation engine
o
Running multiple agents and manage portfolio exposure and risk
Page | 7
Back testing o
Agents are trained to optimize reward or financial quantity such as Sharpe ratio
o
Require out of sample testing to decide if agent qualifies for productive trading
Trading Elements The trading Elements are made up of several blocks the key blocks are the following: Algo Trading Services: The core of the Trading execution is the Algo trading engine. This block allows for the deployment and simultaneous execution of multiple trading strategies on the underlining data which can be made up of live Market Data, other data relevant to trigger conditions and historical data (Market Data and order processing). It also allows the for the capability for back testing for recently developed trading strategies before they go into production. This Service will run under the private cloud infrastructure at the colocation venue and will be deployed per trading customer.
Figure 5: Algo Trading Service
Page | 8
This service provides “Intelligent Order” execution through an external Exchange Broker/HFT Trader based on the different trading strategies which are fed from various Deep learning Trading agents. Market Data Feed Handler: This service also runs on the cloud infrastructure and provides a normalized data market feed into the Algo Trading Service. This can be set up to take any particular market feed and source it into the Algo Trading Engine. Capture and Analytics Appliance(s): This appliance runs within the cloud infrastructure and is managed and provisioned via the private cloud interface. The Appliance is responsible for capturing all raw market data feeds and trading execution. This feeds into a Trading Analytics package that utilizes the Trading Database for a repository. Trading Database: The trading database runs on specialized “Database Elements” of the cloud based infrastructure and is used for fast ingestion of trading analytics data and marketing data and the simultaneous query of the data into the AI Trading Agents and for the back testing of new strategies in the order execution engine. The Database is a clustered configuration and scales in performance and storage based on the number of nodes in the cluster. If offers resiliency from failure and linear scalability in terms of performance by adding additional nodes. Extremely efficient in terms of storage requirements.
Figure 6: Trading Database Cluster
Page | 9
Exchange Broker/HFT Trader: The output of “Intelligent orders” from the Algo trading engine is executed through an external “Exchange Broker” in an automated fashion, The Exchange Broker is an organization authorized to trade directly on the localized exchange. These are companies such as Bloomberg, Interactive Brokers, JPMC, Morgan Stanley etc. that have brokerage businesses. If this was a proprietary HFT/UFT Trader using the BINTI AI features a future phase of the BINTI Service Network will be to directly connect with the prop traders U/HFT trading environment for trade execution an dproviding a truly hybrid system.
Private Cloud Infrastructure Elements
Figure 7: Private Cloud Infrastructure
The cloud infrastructure is made up of different classes of base network infrastructure, with the addition of the following that can be scaled out: Hypervisor Elements: Used for the Algo Trading Services, Market Data Handler services, AI Trading Agents and services. Database Elements: Used for the scale out of Database nodes for Trading Database.
Page | 10
Capture & Analytic Elements: Specialized appliances that are managed as part of the Cloud management and security. GPU Acceleration Elements: Specialized acceleration entities used by AI Trading Agents to accelerate Deep learning and processing. The cloud environment will be placed at the collocation facility (Nasdaq), at KTH (PDC Data Center) and at the Bank (end user). The number of Hypervisor servers will vary based on the services running on the cloud and this will have to be sized later on.
There is essentially:
A couple of routers, for the uplinks and interfaces into the premise environment. Control servers that manage and provision the cloud environment and then Hypervisor severs that run the applications Trading and AI and Database. A redundant Management and Data path to all. It will be a lot of automation built in as the GPU equipment at KTH and at the collocation which will be linked in just like an extension to the hypervisor servers and the storage at the development site. The Time Distribution Network will be a network overlay which again will be managed through this environment which will be needed at the co-location facility. The Capture/Analytic appliances will also be needed at the co-location facility. All this will be sourced and deployed. The bank, the end user, will not require all the additional things and will get away with a minimal set of hypervisor servers
The private cloud infrastructure and software will be deployed at various worldwide locations to build up the necessary infrastructure for the BINTI Trading Service Network. These locations will be:
Exchange Colocation Data Centers: These will be at various colocations locations throughout the world. Starting in Stockholm. SITLAB CPE: On premise equipment will be installed at SITLAB for the secure deployment of AI Trading Agents and updates from Development into Operation. Trading Customer CPE: On premise equipment will be installed at the Trading Customer for the secure flow of trading strategy recommendations from back testing analysis to enable deployment of the trading strategy at that colocation facility. SITLAB Regional and WW AI Centers: These are not trading execution sites but sites that consolidate “Intelligent Trading” data at a regional and WW level. Specialized machine learning Trading agents will be learning about trades across different markets, as well as taking input on geo-political data feeds with the specific task of making Trading recommendations based on a specific colocation that will be fed into the order execution queue at the specific local colocation site.
Page | 11
Ultra-Low Latency Competitive Artificial Intelligence (AI) and/or Machine Learning As for quite a long time, automated trading systems have been used to monitor and act on stock information by using artificial intelligence systems trained with historic trends and market data. These predictions can continuously be honed and improved by circular feeds of response and effectiveness of the recommended actions and their results. Artificial Neural Networks, Bayesian Analysis, Fuzzy Logic and Hidden Markov Chains are commonly deployed in these AI applications. We aim to further develop the AI powered decision-making platform, that makes it possible to perceive and affect the ways in which many agents interact in complex environments. The SITLAB team will comprise of experts in machine learning, probabilistic modelling, Gaussian processes, reinforcement learning, decision theory, multi agent systems and game theory—all of which goes into the company’s aim to build an AI decision making platform on a foundation of interpretable principles of mathematics and learning. Statistical approaches and artificial intelligence (AI) applications are widely used within the field of high-frequency trading (HFT) to automatize the trading process. The overall process of trading and risk assessment requires the ability to extract data from numerous data streams in real time, then analyze this data and take actions based on the analysis.
Figure 8: Competitive Trading Agents in U/HFT Trading
Page | 12
With the Binti Project we take off from prior art and proposes an innovation that allows the AI learning to benefit from a competitive environment. This is achieved by deploying several trading blocks in the system utilizing alternative hardware and software technology including programming language, trading algorithms and machine learning environment. While each trading block uses its own AI to make a decision, the overall architecture allows the AI to consider the decisions of other trading blocks. Moreover, the architecture enables sandboxing, i.e., the simulation of a trading environment. The system can replay specific trading situations and the AIs can compete for the trade position even offline – besides the trading hours – to improve their strategies. Besides the benefit for the training of the algorithms, with the versatility and hardware diversity, computer scientists can explore the best combination of compute and storage hardware to achieve the best possible trading performance for a specific strategy. Furthermore, a certain degree of information sharing between the AIs becomes possible. For example, in the arena of trade order book arbitration and trade result feedback and real time parallel simulation. Success rates and the successful strategies can be shared for the purpose of evolving the AIs and for adapting on rapid changes in the market. This learning compensates for the lack of input to predict the behavior of competing companies trading strategies. The feedback between trading blocks is effectuated at multiple levels, for example: • • •
Choice of algorithm Individual stock level Order book proposal level
On all levels, a monitoring component allows to collect and exchange statistics such as the success rate of a strategy in real time. This enables the overall system to improve the algorithms and instantiate them on the best suitable hardware and software platform. Prior systems for high speed trading have hither too been limited in their trial and error choice evaluation of the best performing trading algorithms. The selection of trading algorithms is an adaptive evaluation process where the adaptation of an algorithm to the trading situation can be quantified as profit. The present innovation allows for a better adaptation to profit through running and evaluating multiple Parallel algorithms simultaneously in parallel architecture and adapting by allowing choice to be effectuated by the best performing trade algorithm according to evaluation of the current trading situation which then selects trades for the order book. By implementing multiple real time trade simulations among which to pick the currently most successful one, an innovative step has been achieved for ultra-fast speed trading that can react upon changes on the stock market. Page | 13
Development and Operational Implementation BINTI deliverable goals The following is the key deliverables/Goals of the BINTI Service Network over the three year period:
BINTI is a Hybrid Enterprise Trading Platform which offer consolidated Ultra High Frequency (UHF)/High Frequency (HF)/Algorithmic Trading (AT) Direct Market Access (DMA) and Overthe-counter Trading (OTC) capabilities which together, in the first stage of full production, will enable enterprise risk management and value enhancement. BINTI will deliver the industry a broker-neutral, quantitatively enriched execution management system (EMS) and encompass all facets of electronic and algorithmic trading The BINTI development project will leverage this open architecture to incorporate the individual strengths of several “best-in-class” tools. The flexibility of this architecture will enable fast, economic growth of the system and the ability to integrate best of breed tools. This project will be carefully structured to ensure the efficient integration of several project partners. Transparency, documentation and a commitment to a hybrid development model are all critical success factors to ensure each project member can contribute to the best of their ability. The BINTI project will follow an agile development model with a series of sprints. These sprints will provide the client with tangible evidence of successful progress toward the ultimate vision and will also provide maximum project flexibility. Prop Trading Desks: o The BINTI platform will be a fully customizable EMS designed to act as a central platform for the creation and execution of algorithmic trading strategies for global equities, futures, options and FX. Over The Counter (OTC): o The BINTI project will provide a unified platform for both the electronic and OTC trading of fixed income instruments. Fund Administration Services: o The BINTI platform will address the unique needs of Administrators which may not be in the market for trade execution purposes but still provide the ability to grow and manage funds.
Page | 14
Sitlab Strategic Technology Partners The following table outlines the key partners involved in the key hardware and software elements and infrastructure providers. Element Development Platform Infrastructure FPGA Hardware Development GPU Hardware Development
Organization
Location
Website
KTH PDC CEI N/A
Stockholm, Sweden Colorado Springs, USA N/A
https://www.pdc.kth.se https://coloradoengineering.com N/A
Flink Flink Nvidia Kinetica RACS/RAVCAP
Switzerland Switzerland Cambridge, UK London, UK Atlanta, USA
N/A N/A www.nvidia.com www.kinetica.com N/A
Vela Trading AlgoTrader Solarflare Quasardb RACS/RAVCAP
London, UK Switzerland Cambridge, UK Paris, France Atlanta, USA
www.tradevela.com www.algotrader.com www.solarflare.com www.Quasardb.net N/A
Private Cloud Elements Private Cloud Infrastructure Private Cloud Deployment Private Cloud Management
CloudSeeds CloudSeeds CloudSeeds
Hamburg, DE Hamburg, DE Hamburg, DE
www.cloudseeds.de www.cloudseeds.de www.cloudseeds.de
Business Development Trading and Cloud Integration AI Integration Binti multi-phase deployment Competitive AI Integration
JAAG Inc. Flink JAAG Inc. RACS/RAVCAP
Los Angeles, CA, USA Switzerland Los Angeles, CA, USA Atlanta, USA
www.jaagnet.com N/A www.jaagnet.com N/A
AI Elements Trading Development Trading Agent/Runtime GPU Acceleration Application GPU Database Game Theory Application Trading Elements Market Data Handler Algo Trading platform Capture & Analytics Trading Database Asset Management Automation
Table 1: Technology Partners
These are the key partners that will be delivering the BINTI trading service network.
Page | 15
BINTI Roll out strategy BINTI will be rolled out in phases the following key phases of the project will be implemented in the approximate time frame: Phase
Go Live Date
AI
Trading
Cloud
Phase1
October 2018
GPU and Cloud based Trading Agents
Exchange Broker Trading Only
Phase 2
March 2019
As Above plus U/HFT Trading interfaces
Phase 3
September 2019
As above plus Regional Trading Agent first release Leverage IBM AI technology at Regional sites As above plus WW Trading Agent First Release
Phase 4
March 2020
As above plus Leverage IBM AI technology at WW sites
As above
Phase 5
October 2020
As above
As above
Phase 6
March 2021
Exit Strategy/IPO
Exit Strategy/IPO
Stockholm Colocation Venue only SITLAB CPE 1 Trading Customer CPE As above plus London and Frankfurt Colocations Frankfurt Regional DC in IBM facilities As above plus NY, Chicago and Dubai Colocation sites. Dubai and Washington DC Regional Sites Stockholm WW site As above plus Chennai, Tokyo, Hong Kong and Singapore colocations plus Tokyo regional site As above plus Sydney, JS, Sao Paulo, Mexico and Canada colocation sites and Singapore Regional site. Exit Strategy/IPO
As above
Table 2: BINTI Rollout Strategy
Page | 16
Cloud based AI Trading Service Network (BINTI) Summary SITLAB has mapped out their Vision and Architecture for a Cloud based AI Trading Service Network that spans the globe known as BINTI. It has highlighted the major elements of the system namely AI, Trading and Private cloud. It has highlighted the primary technology providers and third-party coordinators involved in implementing such a service network. It has provided a timeline of rolling such a platform out worldwide over a three-year period. During this period, it will be developing a vast array of intellectual property and knowledge which will stand it in a good valuation for a possible exit strategy. This IP will be developed in the areas of:
AI Self Learning Trading Agents Distribution of Intelligence across Regional and world wide levels and its use in Trading decisions Integration of Algorithmic based AI trading and U/HFT trading for a very Hybrid environment.
Custom Custom Custom
Custom Custom Custom
Custom Custom Custom
Custom Custom Custom
Colocation Infrastructure Algo
Algo
Algo
Cloud Infrastructure
Intelligent KPIs Algo
IBM Cloud London 02, England
Algo
IBM Cloud Washington, D.C. 01 USA
IBM Cloud Frankfurt 02, Germany
Consolidated AI Data
Custom Custom Custom Algo
Custom Algo Custom
Cloud Infrastructure
Custom
IBM Cloud Chennai 01, India
IBM Cloud Tokyo 02, Japan
IBM Cloud Singapore 01, Singapore
AI Service Deployment & Updates
Intelligent Order
Frankfurt, New York, London, Chicago, Tokyo, Singapore, Hong Kong, Dubai Customs Emulated Trading Platform
Figure 9: BINTI Cloud Based Infrastructure World Wide
Page | 17
Exchanges
Exchange Colocation facilities
AI Service Deployment & Updates SITLAB at KTH (Stockholm)
SITLAB AlgoTrader 3 Cloud Infrastructure
Trading Exchange N. Amer 1.
N. America SITLAB Regional A/I Services
SITLAB AlgoTrader 1 Cloud Infrastructure
SITLAB AlgoTrader 3 Cloud Infrastructure
Trading Exchange EMEA2
WW SITLAB AI Services
SITLAB AlgoTrader 1 Cloud Infrastructure
EMEA SITLAB Regional A/I Services
SITLAB AlgoTrader AlgoTrader 2 2 Cloud Infrastructure
Trading Exchange EMEA1
SITLAB DEV. OPS LAB
SITLAB WW A/I DC
SITLAB AlgoTrader AlgoTrader 1 1 Cloud Infrastructure
Intelligent Trades
SITLAB Cloud Colocation Infrastructure locations
SITLAB Cloud Regional Infrastructure locations
SITLAB Cloud WW & Dev. Infrastructure locations
Figure 10: WW AI Agent and AI KPI Distribution
Exchange Colocation facilities
Exchanges
AI Service Deployment & Updates
SITLAB AlgoTrader 3 Cloud Infrastructure
Trading Exchange N. Amer 1.
Trader 1 CPE
SITLAB AlgoTrader 1 Cloud Infrastructure
SITLAB AlgoTrader 3 Cloud Infrastructure
Trading Exchange EMEA2
Trader 2 CPE
SITLAB AlgoTrader 1 Cloud Infrastructure
SITLAB AlgoTrader AlgoTrader 2 2 Cloud Infrastructure
Trading Exchange EMEA1
Traders Dev Ops locations
Trader 3 CPE
SITLAB AlgoTrader AlgoTrader 1 1 Cloud Infrastructure
Intelligent Trades
SITLAB Cloud Colocation Infrastructure locations
SITLAB Cloud Regional Infrastructure locations
Traders CPE Cloud Infrastructure locations
Figure 11: WW Trader Access
Page | 18
Sitlab core technology will be founded on mathematical principles from three previously segregated fields: probabilistic modeling, machine learning and game theory. By merging these three fields, the Binti platform allows for decision-making based on interpretable principles for the first time ever. The platform will use powerful statistical tools to generate flexible, dynamic probabilistic models which provide new insights about virtual or physical environments; machine learning and decision-making methodologies that are more visible and interpretable than those that take place within deep neural nets; and multi-agent systems that are much more flexible, adaptable and strategically interactive than traditional decision-tree based systems. Sitlab team will comprise experts in machine learning, probabilistic modeling, Gaussian processes, reinforcement learning, decision theory, multi agent systems and game theory—all of which goes into the company’s aim to build an AI decision making platform on a foundation of interpretable principles of mathematics and learning. We aim to further develop the AI powered decision-making platform, that could makes it possible to perceive and affect the ways in which many agents interact in complex environments.
Scinture Innovation Technology Laboratory (Sitlab) AB (Publ) BOX 7177 103 99 Stockholm Sweden
[email protected]
Page | 19