Generative Adversarial Construction of Parallel Portfolios

Abstract

Since automatic algorithm configuration methods have been very effective, recently there is increasing research interest in utilizing them for automatic solver construction, resulting in several notable approaches. For these approaches, a basic assumption is that the given training set could sufficiently represent the target use cases such that the constructed solvers can generalize well. However, such an assumption does not always hold in practice since in some cases, we might only have scarce and biased training data. This article studies effective construction approaches for the parallel algorithm portfolios that are less affected in these cases. Unlike previous approaches, the proposed approach simultaneously considers instance generation and portfolio construction in an adversarial process, in which the aim of the former is to generate instances that are challenging for the current portfolio, while the aim of the latter is to find a new component solver for the portfolio to better solve the newly generated instances. Applied to two widely studied problem domains, that is, the Boolean satisfiability problems (SAT) and the traveling salesman problems (TSPs), the proposed approach identified parallel portfolios with much better generalization than the ones generated by the existing approaches when the training data were scarce and biased. Moreover, it was further demonstrated that the generated portfolios could even rival the state-of-the-art manually designed parallel solvers. Index Terms-Automatic portfolio construction (APC), generative adversarial approach, parallel algorithm portfolio, parameter tuning.

Automatic constraint programming solver selection method based on machine learning for the cable tree wiring problem

“…Furthermore, we randomly sampled instances from the problem space during the model training process to form the training set, assuming that the training instances adequately represent the test instances. While this assumption is ideal, due to the dynamic changes and complexity in data distribution, it is only sometimes accurate, as observed in works by Tang et al [51] and Liu et al [52]. Future research directions should address training-test distribution inconsistency, exploring domain adaptation, transfer learning, meta-learning, and online learning techniques.…”

“…[49] proposed one method named automatic construction of parallel portfolios (ACPP), which automatically constructs effective parallel portfolios based on a given set of problem instances and an extensive design space. Additionally, other works such as Liu et al (2019) [50], Tang et al (2021) [51], and Liu et al (2020) [52] also experimented with the automatic construction of ensembles.…”

“…(2021) [51], and Liu et al. (2020) [52] also experimented with the automatic construction of ensembles.…”

A Survey of Methods for Automated Algorithm Configuration

“…This is related to the cold-start problem in ML, in that users of AC systems may not have gathered enough instances to properly train their parameterized algorithm to solve their problem effectively. While some work has been performed in this direction in terms of automatically generating instances Smith-Miles and Bowly (2015); Akgün et al (2019); Tang et al (2021); Liu et al (2022), there are still many real-world problems that cannot be modeled with these techniques.…”

Few-Shots Parallel Algorithm Portfolio Construction via Co-Evolution

“…Built upon automatic algorithm configuration, the automatic construction of parallel algorithm portfolios (PAPs) [11]- [15] seeks to identify a set of configurations to form a PAP. Each configuration in the PAP is called a component solver.…”