"Keep being coupled to other people, keep spreading your ideas, because the sum of all of us together, coupled, is greater than our own parts." - Hasson U, 2016
We work at the intersection of statistics and AI, with a particular focus on developing principled methods motivated by real-world challenges in industry and science. Recent interests include:
Trustworthy ML
Developing algorithms that generalize across different conditions:
random noise and high frequent signals.
structured noise such as batch effects, background features, and shifting market conditions.
non-stationaries such as time-varying market conditions (e.g., bear or bull market).
Designing algorithms that adapt to the target data or tasks such as a specific sub-task or the most recent market environment (similar to transfer learning but in a more adaptive way).
Proposing adaptive learning methods, including ensemble learning and invariant learning methods.
Efficient GenAI
Discrete diffusion models for efficient sampling and generation.
Multimodal models with a recent focus on unifying language and vision generation and editing.
Alignment for improving generalization in generative models and improving reasoning capabilities for solving complicated tasks.
Foundations of AGI from a statistical perspective
Developing next-generation statistical theory that goes beyond classical large-sample asymptotics and recent measure concentration/high dimensional probability tools.
Unifying frequentist and Bayesian paradigms from the generalization perspective.
AI for tech, finance, and science
Quantitative finance.
Personalized adverting and recommendation systems.
Materials science and engineering.
We no longer have time to run reading seminars. We used to run the following reading seminars:
Fall 2022 and Winter 2023: Learning + X
Winter and Summer 2022: RL + X
Fall 2021: RMT4ML
Winter 2021: RSL and SL
Fall 2020: RA and RL
Fall 2018: High Dimensional Inference
Winter 2018: Statistical Machine Learning
Reading seminar tips (from Stanford ML group)