Selection Hyper-heuristics Can Automatically Adjust the Learning Period to Optimally Solve Pseudo-Boolean Problems

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Doerr, Benjamin, Oliveto, Pietro S., Warwicker, John Alasdair
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917543887765504
author Doerr, Benjamin
Oliveto, Pietro S.
Warwicker, John Alasdair
author_facet Doerr, Benjamin
Oliveto, Pietro S.
Warwicker, John Alasdair
contents The Random Gradient hyper-heuristic was recently shown to be able to learn the optimal neighbourhood size when optimizing the LeadingOnes benchmark via the Randomised Local Search (RLS) meta-heuristic. However, for this to happen, a learning period of a certain length $τ$ had to be used, differently from classic hyper-heuristics, which change their behaviour based on the success of only the previous iteration. In this paper, we show how to automatically set this new parameter value, relieving the user from the non-trivial task of controlling this novel algorithm parameter. We prove that the resulting hyper-heuristic selects the optimal neighbourhood size in a $1-o(1)$ fraction of the iterations and, consequently, optimises the LeadingOnes benchmark in the best possible time (apart from lower-order terms) achievable with these neighborhood sizes.
format Preprint
id arxiv_https___arxiv_org_abs_2605_29916
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Selection Hyper-heuristics Can Automatically Adjust the Learning Period to Optimally Solve Pseudo-Boolean Problems
Doerr, Benjamin
Oliveto, Pietro S.
Warwicker, John Alasdair
Neural and Evolutionary Computing
Artificial Intelligence
Data Structures and Algorithms
Optimization and Control
The Random Gradient hyper-heuristic was recently shown to be able to learn the optimal neighbourhood size when optimizing the LeadingOnes benchmark via the Randomised Local Search (RLS) meta-heuristic. However, for this to happen, a learning period of a certain length $τ$ had to be used, differently from classic hyper-heuristics, which change their behaviour based on the success of only the previous iteration. In this paper, we show how to automatically set this new parameter value, relieving the user from the non-trivial task of controlling this novel algorithm parameter. We prove that the resulting hyper-heuristic selects the optimal neighbourhood size in a $1-o(1)$ fraction of the iterations and, consequently, optimises the LeadingOnes benchmark in the best possible time (apart from lower-order terms) achievable with these neighborhood sizes.
title Selection Hyper-heuristics Can Automatically Adjust the Learning Period to Optimally Solve Pseudo-Boolean Problems
topic Neural and Evolutionary Computing
Artificial Intelligence
Data Structures and Algorithms
Optimization and Control
url https://arxiv.org/abs/2605.29916