Active Learning Techniques for Pomset Recognizers

TL;DR

This paper advances active learning algorithms for pomset recognizers, enabling more efficient inference of models for concurrent programs with partially ordered structures.

Contribution

It introduces a new counter-example analysis, adapts the $L^{}$ algorithm for efficiency, and designs a finite test suite extending the W-method for pomset recognizers.

Findings

Counter-example analysis can be exponentially more efficient.

The adapted $L^{}$ algorithm reduces redundant queries.

A finite test suite ensures general equivalence of pomset recognizers.

Abstract

Pomsets are a promising formalism for concurrent programs based on partially ordered sets. Among this class, series-parallel pomsets admit a convenient linear representation and can be recognized by simple algebraic structures known as pomset recognizers. Active learning consists in inferring a formal model of a recognizable language by asking membership and equivalence queries to a minimally adequate teacher (MAT). We improve existing learning algorithms for pomset recognizers by 1. introducing a new counter-example analysis procedure that is in the best case scenario exponentially more efficient than existing methods 2. adapting the state-of-the-art $L^{\lambda}$ algorithm to minimize the impact of exceedingly verbose counter-examples and remove redundant queries 3. designing a suitable finite test suite that ensures general equivalence between two pomset recognizers by extending the well-known W-method.