42 - A Guide to Research in Assumption-Based Truth Maintenance Systems in Constraint Satisfaction

Abstract

The reason maintenance problem-solving architecture presents a new paradigm for organizing data and control. This architecture provides an improvement over traditional pattern-directed inference systems in the five areas of artificial intelligence (AI) program design. These areas are efficiency, coherency, flexibility, additivity, and extendibility. Research in reason maintenance has directed its attention toward improving reason maintenance algorithms, inference engine control strategies, and the overall reason maintenance system (RMS)-based system architecture, to ensure that these characteristics are not inadvertently lost. The causal effect of work and objectives in these three areas has become isolated from the real performance issues of the problem solver. This chapter proposes a classification framework to direct and link the three RMS research areas to the contribution of overall performance through the five program design characteristics. It also serves to clarify and guide the discussion of efficiency issues with respect to clear definitions and terminologies in the field of assumption-based truth maintenance. The induction of local classifications of assumption-based truth maintenance system (ATMS) control strategies based on ATMS cost and degree of control further establishes the current state and future directions of ATMS research. The analysis and comparison of the three ATMS research areas in the classification framework suggest that the causal strength of control strategy research is greater than that of reason maintenance algorithm research. To improve an ATMS-based constraint satisfaction problem (CSP) problem solver's run-time efficiency, one should focus on improving ATMS control.