StaticHS: A Variant of Reiter’s Hitting Set Tree for Efficient Sequential Diagnosis

Abstract

Sequential Diagnosis methods aim at suggesting a minimal-cost sequence of measurements to identify the root cause of a system failure among the possible fault explanations, called diagnoses. Hitting set algorithms are often used by such methods to precompute a set of diagnoses serving as a decision basis for iterative measurement selection.We show that there are two natural interpretations of the sequential diagnosis problem and argue that (1) existing methods consider only the more general definition of the problem and that (2) tackling the more specific problem might suffice under assumptions commonly met in practice. Thus, we present StaticHS, a novel variant of Reiter’s hitting set tree usable for solving both formulations of the sequential diagnosis problem. Like Reiter’s algorithm, StaticHS is logics- and reasoner-independent and thus generally applicable to various (diagnosis) domains. Theoretical and empirical analyses show the significant superiority of StaticHS to an application of Reiter’s tree in terms of measurement costs when solving both types of sequential diagnosis problems.

How Should I Compute My Candidates? A Taxonomy and Classification of Diagnosis Computation Algorithms

“…Impact: Single-solution techniques can be highly performant as they may use optimizations that harm completeness by manipulating the set of all solutions for the benefit of computational efficiency [10]. To allow for some degree of control over their performance, multiple-solution approaches are sometimes also configurable, e.g., to compute a number of exactly k solutions, to stop after some timeout occurs, to prune a specified part of the search space, or to stop after a predefined number of search iterations have been performed [50,49,54,16,1]. Relationship: Singlesolution methods are usually one-complete (cf.…”

“…the algorithm's performance may (ceteris paribus) have a bearing on the proper choice between stateful and stateless algorithms (see below). Impact: When memory is the more critical resource, e.g., on small or mobile devices, stateless algorithms may be a way to trade more time for less space, whereas, when time is the more critical resource, stateful algorithms may be preferable [52,54]. Relationship: Stateless algorithms are usually (but not always, cf.…”