An ADCSP-Based Non-Monotonic Framework for Medical Diagnosis

Abstract

The ability to reason within a dynamical environment is of crucial importance in Artificial Intelligence. The present paper models nonmonotonic reasoning by means of a DCSP (Dynamic Constraint Satisfaction Problems) framework, taking advantage of the representation facilities of direct argumentation systems. The algorithm presented below applies dynamic backtracking for the approximate computation of the admissible semantics, which was used to define the concept of multiple diagnosis. The final application of our work is a system for medical diagnosis, that models its search space efficiently and dynamically, while confronted with sequential tests. It asserts and rejects beliefs in different component elements of the diagnosed domain following a nonmonotonic schema which is very close to a human expert’s reasoning model.