Zhanxing Zhu
Machine learning researcher.
ECS, University of Southampton
Southampton, SO17 1BJ, UK
Email: z.zhu@soton.ac.uk
I am Associate Professor at Vision, Learning and Control Group (VLC), School of Electrical and Computer Science (ECS), University of Southampton, UK. I am now closely affiliated with UKRI AI Centre for Doctoral Training in AI for Sustainability. Previously I obtained Ph.D on machine learning from School of Informatics, University of Edinburgh, UK.
I have been focusing on machine learning, particularly, deep learning, broadly covering its theory, methodology and application. Together with my students and collaborators, we attempt to rigorously reveal the underlying mechanism of why deep learning works or not, and inspired by our theoretical understanding and empirical observation, we develop robust, fast and generalizable models and algorithms to boost its applicability in various challenging scenarios and interdisciplinary tasks, e.g. ML4Science. More information is shown in my Google Scholar profile.
Research Interests:
- Understanding deep learning theoretically (SGD, BN, Implicit Bias, Knowledge Distillation, Adversarial Training)
- Robust deep learning models in adversarial and continual environments (YOPO, Inversion Attack, Adversarial Invariant Learning, RCL and BOCL)
- Lightweight and fast large language models (LLMs, ongoing)
- Time series modeling and prediction (STGCN, STFGN, Neural Lad, Functional Relation Field)
- Machine learning for science (ML4Science), including climate, biological and healthcare problems.
Ph.D Studentships. I’m interested in supervising motivated students in the area of AI and machine learning, ranging from theory, algorithms and various applications. Please get in touch to discuss the options and potential topics. You can also check out the UKRI AI Centre for Doctoral Training in AI for Sustainability which has opportunities for 70 PhD students in the area of AI and environmental sustainability.
selected publications
- NeurIPSImplicit Bias of (Stochastic) Gradient Descent for Rank-1 Linear Neural NetworkIn Thirty-seventh Conference on Neural Information Processing Systems (NeurIPS) , 2023
- ICLRImplicit Bias of Adversarial Training for Deep Neural NetworksIn International Conference on Learning Representation (ICLR) , 2022
- NeurIPSSpherical motion dynamics: Learning dynamics of normalized neural network using sgd and weight decayAdvances in Neural Information Processing Systems (NeurIPS), 2021
- NeurIPSYou only propagate once: Accelerating adversarial training via maximal principleIn Advances in Neural Information Processing Systems (NeurIPS) , 2019
- ICMLThe Anisotropic Noise in Stochastic Gradient Descent: Its Behavior of Escaping from Sharp Minima and Regularization EffectsIn International Conference on Machine Learning (ICML) , 2019
- IJCAISpatio-temporal graph convolutional neural network: A deep learning framework for traffic forecastingIn International Joint Conference of Artificial Intelligence (IJCAI) , 2018