Talks

Be aware of model capacity when talking about generalization in machine learning at HKUST, Fudan 2025.

One-step full gradient suffices for low-rank fine-tuning, provably and efficiently at UCLA, HKUST, CityU, SJTU, 2025.

Learning with norm-based neural networks: model capacity, function spaces, and computational-statistical gaps at INRIA, Stuttgart, LMU, UW-Madison, 2024.

From kernel methods to neural networks: double descent, function spaces, curse of dimensionality at Gatsby, UCL; University of Warwick (Statistics Seminar, 2024); University of Genoa (MaLGa Seminar, 2024)

Can we avoid robust overfitting in adversarial training? - An approximation viewpoint (regression part) at EPFL, UW-Madison (MLOPT Idea Seminar, 2024)

The role of over-parameterisation in machine learning - the good, the bad, the ugly (from the function space perspective) at New Faculty Highlights, AAAI 2024

On the Convergence of Encoder-only Shallow Transformers at UBC (MILD Seminar 2024)

Robustness in deep learning: the good, the bad, the ugly at KAUST (Rising Stars in AI Symposium 2023)