About Me

Generative AI  ·  Causal ML  ·  Trustworthy ML

I am a final-year PhD candidate in Computing and Information Sciences at Rochester Institute of Technology, advised by Prof. Linwei Wang in the Computational Biomedicine Lab. My research focuses on building reliable deep generative models: improving their robustness and controllability.

My current work explores these themes in image generation:

• Robustness to attribute correlations. I study how attribute correlations in training data limit generative models’ generalization to underrepresented attribute combinations, and develop approaches to address this challenge.
• Controllable and counterfactual generation. I develop robust causal generative models that account for relationships between attributes, aiming for accurate controllable and counterfactual generation even when attributes are strongly related.

Selected Recent Research

View all publications →

ProReGen: Progressive Residual Generation under Attribute Correlations

ICLR 2026  ·  First author

Correlations among attributes in training data can cause generative models to fail on requested attribute combinations that go against those correlations. Inspired by Robinson’s partialling-out transformation from causal inference, ProReGen turns correlated attributes into orthogonal ones and learns progressively: first from abundant correlation-aligned samples, then from limited correlation-conflicting samples, which only need to teach the model the residual. As a model-agnostic framework, ProReGen applies to VAEs, GANs, and diffusion models, improving the correctness of generations for underrepresented attribute combinations without degrading those for common ones.

Paper Code Poster