ICMS 2024 Session

Machine Learning within
Computer Algebra Systems

ICMS 2024 Home Page — Session Home Page

Jack Trainer — Challenges of Reinforcement Learning in Buchberger's Algorithm

Abstract

Previous work has demonstrated that Reinforcement Learning algorithms can be used to discover new strategies for S-Pair selection in Buchberger's algorithm. This previous work focuses on S-Pair selection when the input polynomial system consists of binomials only. Due to limitations of functon-approximation based reinforcement learning and representing polynomials in a way that is compatible with the most effective function approximation methods, the S-Pair selection strategy learned for binomials performs poorly for non-binomial inputs.

This talk will share experience trying to apply reinforcement learning to learn more effective S-Pair selection strategies for computing Groebner bases of polynomial systems arising in geometric rigidity applications. I will explore some of the challenges that this problem presents from a Reinforcement Learning methodology point-of-view and whether Reinforcement Learning could realistically be applied to develop S-pair selection strategies for some of the more challenging inputs to Buchberger's algorithm.