Transfer Learning, Soft Distance-Based Bias, and the Hierarchical BOA

TL;DR

This paper evaluates a transfer learning technique for hierarchical Bayesian optimization, demonstrating its effectiveness across various NP-complete problems and its potential for significant efficiency improvements.

Contribution

It provides empirical validation of a transfer learning method in hBOA across different problem classes and sizes, showing combined techniques can greatly enhance optimization speed.

Findings

Effective transfer learning on MAXSAT, spin glasses, and vertex cover.

Technique works even with previous runs on different problem sizes.

Combining transfer learning with other methods yields near-multiplicative speedups.

Abstract

An automated technique has recently been proposed to transfer learning in the hierarchical Bayesian optimization algorithm (hBOA) based on distance-based statistics. The technique enables practitioners to improve hBOA efficiency by collecting statistics from probabilistic models obtained in previous hBOA runs and using the obtained statistics to bias future hBOA runs on similar problems. The purpose of this paper is threefold: (1) test the technique on several classes of NP-complete problems, including MAXSAT, spin glasses and minimum vertex cover; (2) demonstrate that the technique is effective even when previous runs were done on problems of different size; (3) provide empirical evidence that combining transfer learning with other efficiency enhancement techniques can often yield nearly multiplicative speedups.