Research from the Centre for Intelligent Transport, School of Engineering and Materials Science at Queen Mary University of London (QMUL), presented at the 2026 IEEE World Congress on Computational Intelligence (WCCI 2026) in Maastricht, explores how machine learning can improve optimisation algorithms.

The study, conducted in collaboration with Bangor University, addresses a common challenge in combinatorial optimisation: algorithm parameters are usually fixed or manually tuned, but this often leads to poor performance across different problem instances.

To overcome this, the proposed method uses machine learning to predict good parameter settings based on problem features, allowing the optimisation algorithm to adapt to each instance before the search starts. This is an example of learning-assisted optimisation, where AI is used to enhance traditional optimisation methods.

The approach was tested on the Electric Capacitated Vehicle Routing Problem and showed better performance than standard globally tuned configurations on unseen instances.

The paper, “Instance-Aware Parameter Configuration in Bilevel Late Acceptance Hill Climbing for the Electric Capacitated Vehicle Routing Problem,” was co-authored by Yinghao Qin, Dr Xinwei Wang, Dr Mosab Bazargani (Bangor University), and Dr Jun Chen, and presented by Yinghao Qin at WCCI 2026.

Reflecting on the work, Yinghao Qin said:

"Machine learning can help optimisation algorithms adapt to different problem instances, improving their performance without replacing them. We hope this work contributes to the growing field of learning-assisted optimisation and encourages closer integration between AI and optimisation."

The work highlights Queen Mary’s contributions to computational intelligence and the growing role of AI in optimisation.