|
 |
Applications and Projects
|
|
|
|
|
|
 Most of our deep computing projects are directly motivated by business problems. Some are well known applications that are used to tackle larger problems or that have new and innovative twists. Other applications are in areas that only recently became amenable to computation or are the result of having information that didn't exist before.
The research and projects in this area span many disciplines: from the scaleable hardware to the software that is tuned to take advantage of the scaleable hardware to algorithms that are designed to extract the best performance from the hardware. Diverse activities, such as computational biology and chemistry, data mining and optimization, protein folding, regional weather forecasting and petroleum reservoir modeling, are all covered by the Deep Computing Institute.
|
|
 |
Optimization
Optimization is central to the field of operations research and
management science, where it is used as both a solution tool and a
modeling device. Optimization is often defined as the minimization or
maximization of a function of several variables usually subject to
several equality or inequality constraints. For many optimization
problems arising in industry, the decision variables represent tasks,
or activities, the constraints represent the availability of resources
to perform these tasks, and the objective is to maximize profit..
Optimization continues to develop rapidly. New models
as well as new solution methods are being developed. In addition, new
computing capability is allowing the solution of ever larger,
real-life, optimization problem.
|
|
|
Projects
|
|
|
 Journey Management
|
|
|
|
Airline Optimization
IBM researchers have played a fundamental role in the development of the underlying mathematical optimization technology for airline scheduling problems. In the case of fleet assignment IBM researchers have worked with airlines to exploit the structure of the problem and devised improved integer programming technology. These improvements have enabled the solution of complex and highly constrained fleet assignment problems with several thousand flights and hundreds of aircraft. In the case of crew scheduling, where the challenges are even greater, IBM has pioneered global solution approaches that have resulted in full crew management decision support solutions.
IBM success in solving these vast comprehensive models is largely due to investment in advancing core optimization methods such as those in the Optimization Solutions Library (OSL), the Volume algorithm and the Branch-and-Cut-and-Price (BCP) framework for high performance parallel computer architectures.
Through work with the financial and utility industries IBM researchers have also made significant strides in the area of stochastic optimization, which will likely prove to be very important in future airline models. IBM Research is also using simulation models to help airlines understand the impact that new technologies, such as self-service kiosks, voice-recognition check-in, smart cards, and electronic ticketing, can have on bottlenecks, personnel needs, and customer service levels. |
|
|
Projects |
|
|
Optimization Solutions Library (OSL) |
|
|
|
Underwriting profitability analysis
In partnership with Farmers Group Inc. and the Insurance Industry Solutions Unit, IBM developed a solution for property and casualty (P&C) insurance risk management. This solution motivated and influenced the design of a new probabilistic estimation (ProbE) based data mining framework. ProbE's algorithm is tuned for the P&C risk management problem, and discovers homogenous risk groups in policy and claims data.
|
|
|
Projects |
|
|
Data Abstraction Research |
|
|
|
Pattern discovery in genetics
A wealth of newly acquired information about genes and proteins is now available in public databases. In fact, is not uncommon for life science companies to accumulate data at a rate of several hundred gigabytes to a terabyte per week. This information may provide solutions to many difficult problems, but understanding what is occurring at the genetic level -- which genes are doing what, and how a protein operates -- requires sophisticated approaches to analyzing this immense store of genetic data. |
|
|
Projects |
|
|
Teiresias |
|
|
|
Web searching
IBM scientists are applying deep computing algorithms to mine in the structure and content of the Web. Our research has led to better algorithms for web search, hypertext classification and community filtering, as well as to fundamental new models of random graphs.
|
|
|
|
Personnel scheduling
Striving to apply state of the art algorithmic techniques to real-life problems IBM researchers designed and implemented an assignment engine that assigns service personnel to service calls. This engine is capable of optimizing a large scale scheduling problem in a few minutes, responding to changing conditions and considering a spectrum of factors simultaneously. The engine has been piloted successfully and was elected as the strategic tool for scheduling all maintenance business of IBM Global Services worldwide.
|
|
|
|
Manufacturing planning and scheduling
Manufacturing planning and scheduling problems arising in industry are often very large, and tend to involve complex constraints specifying the manner in which products are to be built or transported. Because a company typically has numerous conflicting goals, such as increasing market share, reducing inventory costs, and maintaining customer service levels, determining the best plan or schedule can require the generation and evaluation of many possible candidates. IBM has developed numerous planning and scheduling models and algorithms for use in our own manufacturing enterprise. We have used techniques ranging from linear programming to asynchronous agents. Recently, some of these methods have been made available to selected IBM customers.
|
|
|
Projects |
|
|
Steel Industry Solutions |
|
|
Asset Management Tool |
|
|
|
Deep Thunder -- fine scale weather modeling
There are many applications for highly focused weather forecasts in specific industries, whose decision making and operations are weather-sensitive
such as local government, transportation, agriculture, broadcast, energy, insurance, etc. each with specific societal and economic benefits.
To evaluate the potential viability of this approach, the IBM Thomas J. Watson Research Center has a project focused on applications of local, high-resolution, short-term weather forecasting. This effort has been dubbed Deep Thunder.
|
|
|
|
Similarity matching in complex model databases
Scientific research in many different fields relies on the comparison of information stored in immense databases. Researchers at IBM's Thomas J. Watson Research Center in New York are opening new horizons in civil and criminal law, medical research, business, and other areas with new combinatorial computer algorithms. Capable of rapidly sorting through millions of bits of information, these algorithms can compare data that is similar but not necessarily identical.
|
|
|
|
Computational biology
If the 20th century was the age of physics, then the 21st century
promises to be the age of biology. New and fundamental advances in
genetic manipulation, biochemistry, and bio-engineering are now for
the first time allowing us to understand and manipulate, although
still in a very primitive way, some of the most intimate biological
machinery. In this context, computers are increasingly becoming
fundamental mining and discovery tools. We expect that, over the next
10 years, computers running new breed of algorithms - still largely
unavailable at the moment - will help automate a significant portion
of the drug and diagnostic tool manufacturing process. Moreover,
advances in life sciences computational techniques will directly
impact a number of other related sectors, from agrochemical research,
to bio-engineered products, to polymers and smart
materials. |
|
|
|
Project
|
|
|
The Computational Biology Center
|
|
|
|
Bioinformatics and pattern discovery
Bioinformatics is an emerging interdisciplinary science that crosses
the boundaries of several fields such as traditional computer
science, molecular biology, mathematics, chemistry, chemical
engineering and other.
One can summarily describe the field as one whose objective is to
accumulate, manage, and make sense of the vast amount of data that is
becoming publicly available through world-wide efforts on a large
variety of living organisms. The ultimate goal is to
understand the relationships among the various components of a
given organism (or class of organisms),
the flow of information in the underlying biological
mechanisms and processes, and the causal relationships between the
DNA composition of an organism and its externally observable
characteristics. Eventually, the exploitation of this accumulated
knowledge will help in practical tasks covering a vast spectrum
of applications.
The work of the Bioinformatics & Pattern Discovery Group focuses
on a number of problems from computational molecular biology and
in this context, a number of algorithms and tools have been developed
to tackle a wide collection of tasks. These include, but are
not limited to, pattern and association discovery in event
streams (DNA, proteins, gene expression time series, etc.),
homology searching in protein/dna databases, multiple sequence
alignment, text mining in biological databases (both structured and
unstructured), functional and structural characterization of proteins
directly from sequence, and many more.
|
|
|
Project
|
|
|
Bioinformatics and pattern discovery
|
|
|
Pattern Discovery Engine
|
|
|
Teiresias server
|
|
|
Download
|
|
|
Teiresias code
|
|
|
|
|
|
Data mining
Business intelligence (BI) is the gathering, managing, and analysis
of data, which is turned into information that drives strategic
decisions regarding markets to enter or products and
services to develop. BI addresses business problems, such as the
management of customer relationships, merchandise and store
operations, supply chain management, financial management, fraud and
risk.
Business Intelligence encompasses a layered set of activities,
starting with data warehousing and data cleansing, creation and
distribution of summary reports from these data warehouses or data
marts, integration with (OLAP) On Line Analytical Processing tools
that allow end-users to "slice and dice" summaries along different
dimensions (i.e., sales by product, by salesperson, by quarter, by
geography), data mining to discover new and actionable information
from data, and optimization.
|
|
|
Projects
|
|
|
Business Intelligence
|
|
|
|
Visualization
Thanks to the advances in communication bandwidth, network
availability and processor speed, more and more data is being
collected and made available. Scientists and business professionals
need better means to understand, assimilate and communicate the
information contained in this data. Visualization has become an
increasingly important tool for this task. IBM Research studies and
develops techniques and applications for Visualization.
Our research expertise includes 3D graphics and geometric algorithms,
software infrastructure for network visualization, visual perception
science and domain dependent visualization solutions in many domains
such as CAD, medical imaging, weather simulation, Internet media,
banking and finance, etc.
We distribute the results of our research through publications and
patents, participation in standards (e.g. MPEG-4), customer
engagements in collaboration with other IBM organizations (e.g.,
banking institutions, automotive manufacturers), participation to in
company wide projects (e.g. HotMedia), and development of products
(e.g. Data Explorer, Diamond, 3DIX).
|
|
|
Projects
|
|
|
Data Explorer
|
|
|
|
|
|
Theory of computing
IBM researchers are very active in basic research in theoretical
computer science, mainly in the fields of approximation algorithms,
hardness of approximation, graph theory, combinatorial optimization,
network design and routing, and scheduling. Papers authored by IBM
researchers are published regularly in leading scientific journals and
conferences.
|
|
|
|
|
|
