Company Profile, Products & Technologies, Target Market

Brief description of proposed R&D project:

Constraint Satisfaction:
http://www.haifa.ibm.com/dept/vst/simulation_vsml_octopus.html
Overview
Octopus provides an umbrella for generic constraint satisfaction problem (CSP) activities within the Verification and Services Technologies (VST) department. Our solutions leverage two state-of-the-art software assets developed in the department: GEC, a systematic constraint solver implementing Maintain-Arc-Consistency algorithm; and Stocs, a stochastic constraint solver. Our department has vast and long-standing expertise in CSP algorithms and modeling. Our aim is to provide value to IBM through the application of constraint solving to various domains and close interaction with the academic community.


Machine Learning:
http://www.haifa.ibm.com/dept/vst/simulation_vsml_ml.html
Overview
The Machine Learning Group specializes in developing algorithms for automatic pattern recognition, prediction, analysis, classification, and learning of structures. We supply both core technologies and machine learning services.

Our core technologies include:
Bayesian networks
Learning and classifying structures
Anomaly detection
Feature selection
Time series analysis
Support Vector Machines (SVM)


Traffic Congestion
IBM deployed a new Road Charging system for the city of Stockholm in 2006. This highly complex system is used to reduce city congestion by applying a toll on cars entering the city. Although at the beginning the system used RFID and optical measures to identify cars, it is now using only optical-based recognition, due to the very high recognition rate and relatively low cost of using cameras on gantries around the city to identify cars entering.   The Haifa Research Lab provided the novel LPR (License Plate Recognition) engine, as one of the backoffice engines (with part of the processing done in the cameras themselves), which processes cars images and produces their license plate identification. The Lab also has other assets in this area, including operator productivity tools that can reduce human operator efforts required to correct recognition results by up to fivefold.


DMS (DISH Media Services)
DMS is a hybrid architecture (J2EE/J2SE) that balances the requirements for system availability on one hand and system performance on the other. The DMS subsystem provides rich media streaming (both capture and on-demand playback), supporting MPEG2-TS (MPEG2 Transport Stream) and RTP/RTCP standard protocols. The DMS can work with a client player that is RTSP compliant for controlling the on-demand playback session. Recorded stream are captured through a multicast IP/Port, while on-demand playback can be done using either multicast or unicast.  


DISH - Distributed Information Services Hub
The Distributed Information Services Hub (DISH) is a collaborative grid of networked services that collects, manages stores, analyzes, and distributes real-time rich-media information and alerts between ad-hoc producers and consumers of information.

DISH increases shared real-time situational awareness:
-    Utilizes the network to connect everyone to everything
-    Brings relevant personalized information to those who need it
-    Renders live and recorded information available in real time

DISH is an infrastructure that provides an end-to-end solution for command and control (CC) applications.. The server side manages the various input sensors and works with a geo-spatial database  which, in addition to the video asset location on the file-system, records the incoming (periodical) metadata that is associated with the video asset.  In a typical situation the video asset comes with a lower bandwidth metadata channel including additional information (e.g., the geo-spatial coordinates of the footprint of the video, that is refreshed every second or so). The metadata saved in the database can be queried at any time by the end-users. A query on a specific location will return a list of the assets that cover the area of interest. Once the user asks to view one of the retrieved assets, the DMS subsystem (above) handles the streaming from the file-system.

Both DMS and DISH take the SOA approach to allow easier integration with 3rd party applications.

Agent Based Modeling
In agent based model, the system is decomposed into communicating independent agents, and the behavior of each agent is modeled separately. In such a model, the behavior of the entire system is emergent, in that it emerges from the behavior of the individual agent. The behavior of each agent can be modeled using formalism such as mathematical models, economic models and game theory. The advantages of agent based modeling are as follows:
Each agent is modeled separately, and is independent of the other agents. Therefore, it is easy to replace an agent in the system with other, more (or less) sophisticated agents.
It is possible to keep many versions of each type of agent. For a system such as the one above, this would enable keeping separate versions of each persona, based on different intelligence estimates and evaluations.


Container Logistics Optimization:
Overview
The repositioning of empty containers is an enormous logistics task, frequently done manually by logistics experts. This technology automates the process, adding a global view to the problem, while optimizing the global costs.
The solution optimization results in a savings of at least 10% in costs, while allowing customers to scale and better manage their global operation.

Taking a real world problem with many parameters and constraints and putting it into a mathematical model of variables and equations is a big challenge. The model must be flexible enough to allow the insertion of new rules and constraints.
Data cleansing takes data from a variety of sources and extracts accurate data from it. This technology is often required for customers that have multiple IT systems. Data cleansing is one of the most complicated and time consuming components of the optimization solution. It involves a large number of variables (hundreds of thousands) and many constraints/equations (tens of thousands).

Desired Profile of R&D Partner and its role in the proposed R&D project:

The IBM Haifa Research lab is looking for a partner who is capable of joining forces with us to bring new technologies to market.  We are looking for highly skilled, well vested companies with a strong marketing, technical and financial background. The company should have expertise in at least one of the following areas:  healthcare and life sciences, discovery, verification technologies, multimedia, active management, information retrieval, programming environments, business transformation, or optimization technologies.