Wei Luo is a machine learning researcher specialising in robust and reliable AI. His work addresses key challenges in AI system robustness and effectiveness, including uncertainty quantification and nonlinear structures in high-dimensional data, through topological and geometric modelling alongside dynamical systems approaches.
Wei Luo is best known for his contributions to robust and reliable AI, particularly in security and healthcare applications. His work has advanced adversarial robustness and backdoor detection, with key contributions of Topological Evolution Dynamics (TED) for AI security, as featured in the IEEE Symposium on Security and Privacy (S&P), a top conference in Security. His research is regularly published in top-tier AI and machine learning conferences, including ICML, AAAI, IJCAI, and S&P.
Wei Luo is an advocate of rigorous methodological standards and proper reporting of machine learning applications, ensuring that AI research has transparent, reproducible, and cross-disciplinary impact. His widely cited first-author paper, "Guidelines for developing and reporting machine learning predictive models in biomedical research," has influenced best practices in AI for healthcare and beyond, with the majority of citations coming from fields outside traditional machine learning.
At ICOLSEI 2025, Wei Luo will give a talk, "Training Strategies in Deep Learning: Implications for Bias, Safety, and Human Learning." In this talk, he will discuss how different training strategies impact deep learning models, including large language models (LLMs), and their consequences on model bias and safety. The talk will explore the mechanisms by which training data, optimisation techniques, and fine-tuning approaches shape model behaviour, often leading to unintended biases and safety risks.
A key theme of the talk is the connection between machine learning training paradigms and human learning processes. By drawing parallels between training design in AI and structured learning in human education, this talk will explore how insights from education design can inform better AI training strategies and vice versa.