| This dissertation explores the idea of assumption-based reasoning in cooperating problem-solving systems. Assumption-based reasoning is concerned with inference from propositions that are assumed to hold true, but which are not necessarily always true. Assumptive propositions are a result of applying general prototypical knowledge such as "Birds fly", "Swedes are blond", and "Women like men", to a particular situation. The importance of this type of reasoning is linked to the ability to form reasonable inferences in situations where imprecise, uncertain, or very little information is available. Cooperative problem-solving puts together two or more problem-solving systems on a loosely-coupled computer architecture to work in cooperation on a single task. Currently, no practical methods for assumption-based reasoning (or any kind of non-monotonic or non-standard reasoning) exist for multi-agent computing environments. It is realized here using a production system model of computation for the individual agent.; This dissertation develops a framework for building cooperative problem-solving networks of assumption-based reasoning systems. The framework includes, BBRL, the Belief-Based Rule Language, and DATMS, a Truth Maintenance System for cooperative problem-solving. The Multi-Agent Test Environment, MATE, assists in the design, testing, and evaluation of a distributed problem-solving system.; This work was originally motivated by tile need to develop automated data processing techniques for the interpretation of seismic data for the verification of a Comprehensive or Low-Yield Test Ban Treaty. In this problem, seismologists use assumption-based reasoning techniques and collaborate to interpret data from a wide spread network of seismic sensor stations. A prototype interpretation system for this domain forms the basis for experimentation. The design of the system is based on interviews with treaty verification and seismic experts, the single-agent expert system that currently interprets data from the NORESS seismic array in Norway, and specifications developed by the Group of Scientific Experts for the Conference on Disarmament.; Using MATE, we experimented on three communication strategies: broadcast, undirected-request, and directed-request, for networks of size 5, 7, 11 and 15. We also explored the use of communication histories to guide belief revision messages, and the use of introspection to control inference. |
