Multiagent Supply-Chain Coordination and Decision Processes
Speaker
Abstract
| The need for decision support is strongly driven by combinatorial complexity - problems that present decision makers with many options having complex interdependencies. A classic example is coordinating decisions across a multi-tier supply chain. Optimizing such an operation requires close coordination among procurement, inventory management, production scheduling, sales, and fulfillment. Most organizations use a variety of heuristics to decouple these problems, such as monthly or quarterly sales forecasts. |
| Artificial Intelligence researchers have long focused on areas where people seem able to deal with high-complexity problems using a variety of suboptimal and satisfying approaches. Research in AI often involves building computer programs, on the assumption that if one can write a program, one can make a reasonable claim to understanding the problem addressed by the program. Economists are also interested in how people make decisions, with a focus on rational behavior. AI researchers have been drawn to problems in economics precisely because many rational decision problems are clearly too complex for human decision-makers (or even the most powerful computers) to solve. |
| We are interested in supply-chain decision problems both from a business and from an artificial intelligence standpoint. We use the Trading Agent Competition for Supply Chain Management (TAC SCM) as a test bed, because it is a competitive simulation environment that achieves a good balance between real-world complexity and research-grade observability. After seven years of intense international competition, we have a variety of working examples of autonomous agents that approach their decision problems in substantially different ways. This talk will focus on the problem of coordinating decisions across the supply chain, and will highlight several approaches that have been used successfully by some of the most competitive agents that have been developed so far. |
| Contact information: |
| Wolf Ketter |
