ART-AI Seminar & AI / ML group seminar
We are pleased to have Dr. Song Liu, Associate Professor in Data Science and AI in the University of Bristol’s School of Mathematics, join us for this joint ART-AI Seminar & AI / ML group seminar. This will take place on the 22nd April 2026, 15:15pm-16:05pm (GMT) in room 1WN 3.11. If you would like to join this seminar on Teams, please e-mail [email protected]
Abstract: This paper introduces a novel iterative method for missing data imputation that sequentially reduces the mutual information between data and the corresponding missingness mask. Inspired by GAN-based approaches that train generators to decrease the predictability of missingness patterns, our method explicitly targets this reduction in mutual information. Specifically, our algorithm iteratively minimizes the KL divergence between the joint distribution of the imputed data and missingness mask, and the product of their marginals from the previous iteration. We show that the optimal imputation under this framework can be achieved by solving an ODE whose velocity field minimizes a rectified flow training objective. We further illustrate that some existing imputation techniques can be interpreted as approximate special cases of our mutual-information-reducing framework. Comprehensive experiments on synthetic and real-world datasets validate the efficacy of our proposed approach, demonstrating its superior imputation performance. Our implementation is available.
Speaker Info: Song Liu is an Associate Professor in Data Science and AI in the University of Bristol’s School of Mathematics. His focus is on statistical machine learning, and he has worked on a variety of topics including density ratio estimation, missing data, and flow matching / score matching. Prior to his current time in Bristol (where he previously did his masters degree), he worked at the Institute of Statistical Mathematics, Tokyo, and the Tokyo Institute of Technology, where he also completed his PhD.
