Miqing Li

I am an Associate Professor in the School of Computer Science at the University of Birmingham, UK. My research sits at the intersection between AI and optimisation, where I am interested in developing computational intelligence techniques (e.g., evolutionary algorithms) to solve both fundamental and practical optimisation problems. My current research interests include:

Multi-objective combinatorial optimisation;
Expensive and robust optimisation (EAs and Bayesian);
Many-objective optimisation (algorithms, visualisation, assisted decision-making);
General multi-objective optimisation issues (e.g. performance assessment, archiving, fitness landscape analysis and visualisation);
Practical applications (e.g. in software engineering, mechanical engineering, chemical engineering).

This website is my research portfolio where you can find resources of research projects I am/was working on. This includes brief descriptions (ideas and results) of some of my projects on basic research and applied research and their resulting papers (with code and data).

Recent News

June 2024 - Looking forward to presenting our work on visualising landscape of multi-modal multi-objective problems at the PPSN workshop on Multimodal Multiobjective Optimisation. A thank-you to Prof Heike Trautmann for the invitation!
June 2024 - Our work on performing a stepwise adjustment of weight vectors for MOEA/D has been accepted by Swarm and Evolutionary Computation. Congratulations to Xiaofeng Han!
May 2024 - Our work on visualising landscape of multi-objective optimisation problems has been accepted by PPSN'24. Congratulations to Zimin and Zhiji!
May 2024 - Our work (collaborated with Dr Chao Qian's group) on analysing running time of SPEA2 and general elitist MOEAs has been accepted by PPSN'24. Congratulations to Shengjie, Chao and Chao!
April 2024 - Our work (collaborated with Dr Chao Qian's group) on using an archive can bring provable speed-ups in MOEAs has been accepted by IJCAI'24. Congratulations to Chao, Shengjie, and Chao!
April 2024 - Our work (collaborated with Dr Chao Qian's group) on analysing the benefit of diversity maintenance in EAs for multimodal (multiobjective) optimisation problems has been accepted by IJCAI'24. Congratulations to Shengjie, Zhijia, Chao and Chao!
April 2024 - Extension of our FSE'21 and FSE'24 work for multiobjectivising software configuration tuning has been accepted by IEEE Transactions on Software Engineering. Congratulations to Pengzhou and Tao!
March 2024 - Our work on empirical comparison between MOEAs and local search on multi-objective combinatorial problems has been accepted by GECCO'24. Congratulation to Xiaofeng, Xiaochen and Zimin!
March 2024 - Our work (Collaborated with Dr Shuo Wang's group and colleagues in the School of Maths) on multi-objective optimisation for flexible building space usage has been accepted by IEEE CAI'24. Congratulation to Shuo and colleagues!
March 2024 - Our work (collaborated with Prof Guofu Zhang) on estimating and utilising the bounds of feasible solution space for a very constrained multi-objective optimisation problem, test resource allocation of software testing, has been accepted by ACM Transactions on Software Engineering and Methodology. Congratulations to Guofu!
March 2024 - It was my great pleasure to give a talk on MOEAs for combinatorial optimisation at the University of Exeter. A thank-you to Prof Jonathan Fieldsend for the invitation!
January 2024 - Our work (collaborated with Dr Tao Chen) on adapting a bi-objective optimisation model for software configuration tuning has been accepted by FSE'24. Congratulations to Tao!
January 2024 - Our work (collaborated with Dr Liangli Zhen and A STAR* colleagues) on applying multi-objective optimisation to generating neural network architectures with multiple scales has been accepted by IEEE Transactions on Evolutionary Computation. Congratulations to Liangli!

Highlights

[GECCO24] found that 1) MOEAs can compete with local search in combinatorial problems, and 2) a very simple search heuristic, SEMO, performs better than well-established MOEAs (NSGA-II, MOEA/D and SMS-EMOA) and local search algorithms (PLS and anytime PLS) [PDF]

[TEVC23] A systematic survey of archiving methods in multi-objective optimisation from a general theoretical perspective, in which four classes of archiving methods are identified. We also present that archiving methods based on weakly Pareto compliant indicators (e.g., ε-indicator, R2, IGD+) can achieve the same theoretical desirables as archiving methods based on Pareto compliant indicators (e.g., hypervolume indicator).

[IJCAI23] theoretically proved that introducing randomness into the population update process in MOEAs can be beneficial. This finding is against all mainstream MOEAs, challenging their greedy, deterministic population update mechanisms during the search [PDF]

[GECCO23] found that MOEAs are stuck in a different area at a time; but this only happens to combinatorial problems, but not continuous problems [PDF]

[NeurIPS22] A multi-agent dynamic algorithm configuration, with one agent working for one type of configuration hyperparameter, and its instantiation to dynamically configure MOEA/D (e.g., the weights, neighbourhood size, etc) during the search. [PDF]

[TOSEM22] Converting a multi-objective optimisation problem into a single-objective one may not be a good option even if clear preferences between the objectives exist. [Read More]

[TELO21] Through a set of artificial sequences, this work refutes a common belief and shows Pareto-based, indicator-based and decomposition-based criteria may fail on extremely simple Pareto front shapes (i.e. 1D/2D simplex). [Read More]

[FSE21] A bi-objective model to optimise a performance attribute (single-objective) in the software configuration tuning task - a black-box, expensive optimisation problem with rugged landscape and numerous local optima. [Read More]

[TSE20] A critical review and methodological guidance of how to select and use evaluation methods in different multi-objective SBSE scenarios. [Read More]

[ECJ20] An adaptive weight update approach that enables the decomposition-based EMO method to work well for any Pareto front shape. [PDF] [Read More]

[TOSEM20] A customised evolution strategy to generate test suites for software product lines through simultaneously optimising 9 objectives relating to test selection and test prioritisation. [PDF] [Read More]

[CSUR19] A survey of 100 quality indicators in multi-objective optimisation, providing guidance on how to select and use quality indicators in various situations. [PDF] [Read More]

[TEVC18, CEC14] An award-winning work that designs a test problem suite to aid the visual examination of algorithm performance in many-objective optimisation. [PDF1] [PDF2] [Read More]