I am interested in theoretical problems that connect optimization, learning theory, online learning, and modern deep neural networks. My recent work focuses on fine-grained analyses of training dynamics, step-size conditions, Hessian spectral phenomena, and online multi-armed bandit problems with combinatorial structure and budget constraints.

• Optimization behavior of stochastic gradient descent under ill-conditioned loss landscapes, structured Hessian spectra, and architecture-dependent training dynamics.
• Theoretically interpretable perspectives on generalization, optimization stability, and neural network training dynamics.
• Traditional optimization and online learning problems, including combinatorial multi-armed bandits, bandits with knapsacks, and online shortest-path routing.

Outside research, I enjoy Go, good hotels, and discovering good food, even though I am still very far from being a good cook myself. I also like playing Teamfight Tactics and chatting about games, travel, and everyday ideas that make life more interesting.

If you would like to play together, talk about research, or just have a friendly conversation, please feel free to reach out by email: bxl307@student.bham.ac.uk

CV X / Twitter: @xuan10210860 Google Scholar