Prof. Yi Mei

Prof. Yi Mei

Professor, Fellow of Engineering New Zealand, IEEE Senior Member

Machine Learning and Automated Algorithm Design for Optimisation

Designing a good optimisation algorithm usually takes experts years of trial and error, and the result rarely transfers to new problems. My research uses machine learning to automate this: learning which search operator to apply, where to search, how to start, and ultimately generating whole algorithms automatically.

110+papers in this theme
3award-winning papers
7papers with 100+ citations (OpenAlex)
TPAMIlearning-aided neighbourhood search, 2025

Introduction

Learning inside the optimiser

Every metaheuristic is built from components — how to initialise, which operator to apply, which neighbourhood to search, when to accept a move. Each is a decision that machine learning can make better than a fixed hand-designed choice, by learning from experience on many problem instances.

Where machine learning plugs into an optimiserLearned components for initialisation, operator selection, neighbourhood search and acceptance.THE OPTIMISER'S LOOPInitialiselearned initialisationheuristicChoose operatoradaptive operatorselectionSearch neighbourhoodlearned / neuralneighbourhood searchAccept & updateRL-guidedpoliciesAutomated algorithm design: generate the whole optimiser (GP, lifelong optimiser generation)
Figure 1. Where learning plugs into an optimiser. Learned components (gold) replace hand-designed choices; at the highest level, automated algorithm design generates the whole optimiser.

Two complementary learners

Deep and reinforcement learning learn flexible policies from data, while genetic programming learns compact, interpretable components. We study both and how they combine.

My Contributions

With my students and collaborators, I have made the following contributions. The papers shown are selected automatically: award-winning papers, papers with 100+ citations, and papers in top venues. Click a title to read the paper, or list entry to see it in my publication list.

1 Adaptive operator selection

Which operator works best changes during the search and across instances. We learn operator selectors from both offline and online experience, and surveyed the field.

Adaptive operator selectionA selector picks which operator to apply from the search state and learns from the reward.Search state(features)Selector(learned from experience)operator 1operator 2operator 3operator 4newsolutionreward = improvement → update selector
Figure 2. Adaptive operator selection: a selector chooses the next operator from features of the search state and is updated by the improvement it produces.

2 Learning-guided and neural search for routing

We combine neural networks with neighbourhood search and design neural solvers that generalise better across problem sizes and variants.

3 Automated design of algorithm components

GP and learning can design the components of other algorithms: initialisation heuristics, guided local search penalties, and ant colony state transition rules.

4 Combining evolutionary computation and reinforcement learning

We compare and hybridise GP and deep reinforcement learning for sequential decisions, and embed reinforcement learning inside evolutionary search.

Research Output

Papers per yearStacked bars of journal and conference papers per year.05101520252014: 0 journal2014: 1 conference120142015: 0 journal2015: 0 conference20152016: 1 journal2016: 2 conference320162017: 0 journal2017: 6 conference620172018: 1 journal2018: 8 conference920182019: 2 journal2019: 8 conference1020192020: 1 journal2020: 9 conference1020202021: 4 journal2021: 5 conference920212022: 3 journal2022: 6 conference920222023: 3 journal2023: 5 conference820232024: 3 journal2024: 8 conference1120242025: 14 journal2025: 8 conference2220252026: 6 journal2026: 8 conference142026Journal articles & booksConference papers
Figure 3. My papers in this theme per year (112 in total), identified by title keywords in my publication lists; papers can belong to more than one theme. The 2026 count covers papers published so far this year.

Research Themes

Full publication list →   ← Back to home