Research Seminars
2025 — Diffusion Models
A lecture and reading series working through the foundations of diffusion models, from variational autoencoders to score-based generative modelling with stochastic differential equations. Slides, notes, and code are on GitHub: baojian/diff-course.
| Lecture | Topic |
|---|---|
| 01 | Introduction |
| 02 | Variational autoencoders and DDPM |
| 03 | Energy-based models and noise-conditional score networks |
| 04 | Score-based generative modelling through SDEs |
Fall 2024 — Transformers and In-Context Learning
Members: Ziyang Ma, Jialin Wang, Jinzi Luo, Binbin Huang, Yuxiang Wang, Zichen Yu, Silong Wang, Baojian Zhou.
| Date | Location | Paper |
|---|---|---|
| 2024-09-06, 2:00–3:30pm | S301, Zibin Building | Transformers learn in-context by gradient descent |
| 2024-09-13, 2:00–3:30pm | S308, Zibin Building | Linear Transformers are Versatile In-Context Learners |
| 2024-09-20, 2:00–3:30pm | S308, Zibin Building | The emergence of clusters in self-attention dynamics |
| 2024-09-27, 2:00–3:30pm | S308, Zibin Building | A mathematical perspective on Transformers |