Research Area

Operations Research is practiced throughout IBM, with the Research Division providing a focal point for activities in the science and the application of this field. Research and algorithm development continues in the areas of optimization, statistics, queueing theory and agent-based systems. Current application areas include vehicle routing and staffing, supply chain modeling and optimization, manufacturing planning and scheduling, transportation modeling, process industry scheduling, service industry resource planning and scheduling, airline optimization and forecasting. New initiatives include modeling and software environments for collaborative planning and electronic commerce.

Almaden Research Center

At Almaden Research Center, the Operations Research-related work has been in:

• Research in algorithms, complexity, and machine learning done by members of the Principles and Methodologies Departments ("The Theory Group").
• Research done by members of the Storage Systems function; in particular the following projects: Problem Isolation - Gestalt Analysis of Unstructured Logs (GAUL), Proportional Adaptation for Conservation of Energy (PACE), and the Quality of Service project.
• Research done by members of Almaden Services Research function, in particular Complex Service Systems Modeling and Simulation and Service Quality and Experience.

Dublin Research Lab

The focus of Operations Research activities at the IBM Research and Development Lab in Dublin, Ireland, is on planning under uncertainty and nonlinear optimization. Our work is motivated by real-world challenges arising from problems in water, energy, and transport. Urban systems involve some amount of uncertainty in demand, price, supply, and costs, which is often the consequence of other factors such as human behavior and preferences, natural phenomena (for example, weather or natural disasters), and economic and societal changes. Moreover, large-scale data sets collected via sensors deployed in urban systems are often noisy and also contribute to the uncertainty. When developing models for decision support, it is therefore crucial to find solutions that take into account uncertainty. It is also essential to provide information on uncertainty to real-world decision makers in a way that is intuitive and understandable, without requiring prior knowledge of mathematical optimization.

Therefore, we focus on both mathematical optimization techniques, and techniques for making dealing with uncertainty more accessible to real-world decision makers. Some of the mathematical techniques we consider include stochastic optimization, robust optimization, scenario-based hedging, constraint programming and distributionally robust optimization.

To bring the power of these techniques to real-world decision makers, we are investigating graphical user interfaces that best communicate uncertainties and trade-offs. In addition, models of urban systems involve highly nonlinear dependencies which must be taken into account in order to obtain accurate and reliable optimal solutions. To address this challenge, our experts in nonlinear optimization are developing new methods in areas such as mixed integer nonlinear optimization, convex optimization, semi-definite programming, multi-objective optimization and constraint programming.

We are currently applying our research to real world problems such as design and operations planning of water supply and distribution systems and real-time planning of traffic systems.

Haifa Research Lab

Most of the Operations Research related research in HRL is done by two groups: the Constraint Satisfaction (CSP) group (together with the Formal Verification group) and the Business Optimization group. Those groups deploy several OR methods and their combinations (LP, MIP, CSP, SAT, Statistical technics) using IBM generic solvers such as GEC (systematic CSP solver), STOCS (stochastic CSP solver), MAGE (SAT solver), ILOG CPLEX (LP, MIP solver), ILOG CP Optimizer (CSP solver).

With professional skills in both Operation Research and Computer Sciences, these groups focus on developing novel algorithms and tools for almost all aspects of interest to IBM. We solve real-life problems ranging from the domain of IBM's smarter planet initiative such as travel, transportation and water systems, through the domain of business improvement such as call centers, logistics, business processes, work force management and systems engineering to the domain of computer design and services such as floor planning, configuration management, formal hardware verification, simulation-based verification and cloud computing. More specifically, our tools and technology are at the core of verification of all high-end hardware designed by IBM, we are at the core of planning and scheduling IBM's GBS ~150K professional services workforce, as well as provide call-center scheduling applications. In addition, we provide technology at the core of large IBM's business solutions engagements, in particular in configuration and optimization.

Our research interests include (but are not restricted to) systematic and stochastic solving of constrained optimization problems, hybrid approaches, solving large scale problems, soft constraints and preferences, multi-objective optimization, conflict analysis, SAT algorithms, conditional CSP's, modeling (including language expressiveness, type definitions, and platforms), simulation and forecasting.

Last but not least, we are highly involved in deep two-way relations with academia: collaborating, organizing workshops, presenting invited talks and tutorials at major conferences, editing journal issues, writing papers and book chapters, and more. Our paper describing the work force management of services professionals won the best application paper award at CP'10. We had four lecturers and two sessions chairs at ORSIS (Operation Research Society for ISrael) 2011 conference.

Tokyo Research Lab

The Operations Research team at IBM Research - Tokyo faces increasingly more needs and opportunities in working closely with researchers in other areas to develop complex solutions. A primary responsibility of the OR team is to develop key algorithms that can differentiate such solutions. For example, the OR team is contributing to the project of developing IBM Mega Traffic Simulator. A primary role of the OR team in this project is in developing the algorithms for searching the path that a driver in the simulator takes, where the major issues include speed of the algorithm and how risk is taken into account. The scalability of the simulator heavily depends on the speed of the path-search algorithm, and the behavior of the simulator largely depends on how risk is taken into account in the path-search algorithm. Other notable contributions of the Operations Research team are in machine learning, where scalable algorithms are needed to deal with a large amount of data.

The Operations Research team has also been developing advanced optimization techniques for clients of IBM. In particular, for over ten years, we have been developing Production Design and Operations Scheduling (PDOS) solution for steel manufacturers in collaboration with IBM Research - Watson Research Center. There has also been a focus on optimization algorithms for logistics.

Watson Research Center

The Business Analytics and Mathematical Science Department provides mathematical support for business functions throughout IBM and within Research strategies. We address the fundamental observation that business metrics are quantitative and businesses are only competitively advantaged when they optimize key metrics. The work includes optimization of processes with focus on efficiency in resources, quality, risk management, costs/profits, pricing, supply chains, etc. The department also leverages IBM Research's skills in applied probability, optimization, simulation, statistics, queueing, and related areas to drive better business decision making within IBM and, in partnership with GBS, for our clients. This work is partitioned by industry and application domain. We also do fundamental, foundational research in a number of areas, including cryptography, security, information retrieval, complexity theory, data mining, database principles, distributed systems, systems management, online algorithms, geometric algorithms, statistical physics, computational biology, approximation algorithms, queuing theory, combinatorial optimization, learning theory, descriptive complexity, stochastic optimization, derivative-free optimization, constraint programming, SAT, and automatic algorithm configuration. Our activities are organized around three disciplinary groups and two application groups:

• The Analytics Architecture group is focused in software engineering issues related to Business Analytics and is in charge of coordinating the department's activities in the Open Source space.
• The Data Analytics Center is focused on research in data mining, predictive modelling, risk analysis and statistical analysis and forecasting.
• The Optimization Center is focused on research in algorithms and theoretical Computer Science, mathematical programming, applied mathematics and high performance computing in sequential and parallel models.
• The mission of the Services Modeling group is to apply our deep technical expertise in disciplines such as optimization, forecasting, and probabilistic analysis to the areas of Business Analytics and Workforce Management, where we develop solutions that include services project/portfolio management, skill analytics, demand forecasting, workplace learning and optimization, and strategic planning.
• The mission of the Production Modeling group is to apply our deep technical expertise to the areas of Supply Chain Analytics, Production Planning and Modeling, and Petroleum and Energy Analytics, where we develop solutions that include steel mill production planning and inventory management, demand conditioning, supply chain simulation, and reservoir modelling. A new focus of this group is the exploitation of artificial intelligence methods for optimization.