Approximate Bayesian Inference for Science

Opportunities and barriers at the interface between machine learning and science. Disentanglement, decision-making, interpretability, etc.

Deep Generative Models for Single-cell Transcriptomics

We develop a collection of DGMs for principled, scalable and fully-probabilistic analysis of multiple scRNA-seq datasets. Examples include scVI, scANVI, gimVI and totalVI.

Structured Counterfactual Inference

Several practical instances of batch learning from bandit feedback with added structure. Collaborations with Ant Financial Services & Inc.