Machine Learning for Online Algorithm Selection under Censored Feedback

TL;DR

This paper addresses the challenge of selecting algorithms in online settings with censored runtime data, proposing adapted bandit algorithms that improve selection performance using censored feedback.

Contribution

It introduces adaptations of multi-armed bandit algorithms for online algorithm selection with censored runtime data, maintaining efficiency and improving performance.

Findings

Thompson sampling-based methods perform strongly in experiments.

Adapted algorithms effectively handle censored runtime data.

Proposed methods outperform existing approaches.

Abstract

In online algorithm selection (OAS), instances of an algorithmic problem class are presented to an agent one after another, and the agent has to quickly select a presumably best algorithm from a fixed set of candidate algorithms. For decision problems such as satisfiability (SAT), quality typically refers to the algorithm's runtime. As the latter is known to exhibit a heavy-tail distribution, an algorithm is normally stopped when exceeding a predefined upper time limit. As a consequence, machine learning methods used to optimize an algorithm selection strategy in a data-driven manner need to deal with right-censored samples, a problem that has received little attention in the literature so far. In this work, we revisit multi-armed bandit algorithms for OAS and discuss their capability of dealing with the problem. Moreover, we adapt them towards runtime-oriented losses, allowing for partially censored data while keeping a space- and time-complexity independent of the time horizon. In an extensive experimental evaluation on an adapted version of the ASlib benchmark, we demonstrate that theoretically well-founded methods based on Thompson sampling perform specifically strong and improve in comparison to existing methods.