Yongtao Wu (2026) ‒ LIONS ‐ EPFL

Research Interests

Deep learning
Over-parametrized models

Biography

On August 24th, 2026 I successfully defended my PhD thesis. The thesis, entitled “Optimization in Modern Machine Learning: Steepest Descent Theory and Trustworthy Models” was supervised by Professor Volkan Cevher.

I received a Bachelor’s degree in Telecommunication Engineering from Sun Yat-sen University in 2020, and a Master’s degree in Machine Learning from KTH Royal Institute of Technology in 2022. From July to September 2022 I was an intern at LIONS. In September 2022 I started my PhD there.

Diffusion-based Cumulative Adversarial Purification for Vision Language Models

J. Fu; Y. Wu; Y. Chen; K. Peng; X. Zhang et al.

Transactions on Machine Learning Research. 2026. Vol. 2026-June.

Research Interests

Biography

Diffusion-based Cumulative Adversarial Purification for Vision Language Models

Optimization in Modern Machine Learning: Steepest Descent Theory and Trustworthy Models

Linear Attention for Efficient Bidirectional Sequence Modeling

Robustness in Both Domains: CLIP Needs a Robust Text Encoder

Multi-Step Alignment as Markov Games: An Optimistic Online Mirror Descent Approach with Convergence Guarantees

Machine learning-aided hysteretic response prediction of double skin composite wall under earthquake loads

Quantum-Peft: Ultra Parameter-Efficient Fine-Tuning

Single-pass Detection of Jailbreaking Input in Large Language Models

Hadamard product in deep learning: Introduction, Advances and Challenges

Membership Inference Attacks against Large Vision-Language Models

Universal Gradient Methods for Stochastic Convex Optimization

Revisiting Character-level Adversarial Attacks for Language Models

Robust NAS under adversarial training: benchmark, theory, and beyond

On the Convergence of Encoder-only Shallow Transformers

Extrapolation and Spectral Bias of Neural Nets with Hadamard Product: a Polynomial Net Study