DynamicHS: Streamlining Reiter’s Hitting-Set Tree for Sequential Diagnosis

Given a system that does not work as expected, sequential diagnosis aims at suggesting series of measurements to isolate the true explanation for system’s misbehavior from potentially large set possible explanations. To reason about best next measurement, methods usually require sample fault explanations each step iterative diagnostic process. The computation this can be accomplished by various search algorithms. Among those, Reiter’s HS-Tree is one most popular due its desirable properties and general applicability. Usually, used in stateless fashion throughout process (re)compute per iteration, time given latest (updated) knowledge including all so-far collected measurements. At this, built tree discarded between two iterations, albeit often parts have rebuilt involving redundant operations calls costly reasoning services. As remedy we propose DynamicHS, variant maintains state session embraces special strategies minimize number expensive reasoner invocations. DynamicHS provides an answer longstanding question posed Raymond Reiter his seminal paper 1987, where he wondered if there reasonable strategy reuse existing compute after new information obtained. We conducted extensive evaluations on real-world problems domain knowledge-based systems—a field usage state-of-the-art—under scenarios terms computed heuristic measurement selection used. results prove reasonability novel approach testify clear superiority wrt. time. More specifically: (1) required less than 96 % executed sessions. (2) exhibited substantial statistically significant savings over scenarios, with median maximal 52 75 %, respectively. (3) relative amount saved appears neither depend nor heuristic. (4) In hardest (most time-intensive) cases scenario, achieved even higher average, could avoid overheads 175 800 respectively, opposed HS-Tree. Remarkably, achieves these performance improvements while preserving well applicability