CSE DSI Machine Learning Seminar with Jiawei Zhang (CS, UW Madison)
Learning and Generation Under Constraints: New Results on Constrained Optimization, Bilevel Optimization, and RLHF
Many machine learning and engineering problems involve constrained learning and generation, such as distributed learning, reinforcement learning, resource allocation, and alignment problems. Bilevel optimization is also a common constrained optimization problem, appearing in meta-learning, neural network architecture search, and data mixture optimization.
In the first half of this talk, we will discuss the latest first-order algorithms for solving constrained optimization problems, including optimization with stochastic constraints and bilevel problems where the lower-level problem has constraints or is non-strongly convex. In the second half, I will present recent results on alignment and introduce a more robust alignment algorithm that better accommodates the preferences of different user groups.
Jiawei Zhang is an incoming assistant professor in the Department of Computer Sciences at the University of Wisconsin-Madison. Currently, he is a postdoctoral fellow in the Laboratory for Information & Decision Systems (LIDS) at MIT, working with Prof. Asuman Ozdaglar and Prof. Saurabh Amin. He obtained his PhD in Computer and Information Engineering from the Chinese University of Hong Kong, Shenzhen, under the supervision of Prof. Zhi- Quan (Tom) Luo. Previously, he earned his BSc in Mathematics (Hua Loo-Keng Talent Program) from the University of Science and Technology of China.
His research interests include:
- Optimization theory and algorithms with applications in machine learning, energy, and signal processing
- Optimization, generalization, and robustness of machine learning, reinforcement learning, generative models (including diffusion models, large models, foundation models)
Keller 3-180 or via Zoom.