How should we represent various types of concepts?
How to capture the compositional and causal structures underlying these concepts?
These research areas share a unified goal: building interpretable, identifiable representations
Theoretical foundations and identifiability guarantees
Causal Representation Learning:
Counterfactual Reasoning:
Controllable generative models guided by causal principles and identification guarantees
Multi-Domain Image Generation:
Unpaired Image Translation:
Efficient Generation:
Extending identifiability to multimodal representations
Modular Vision-Language Alignment:
Controllable Generation with Concepts:
Text-Guided Image Manipulation:
Optimizing and interpreting trained generative models
Interpretability of Generative Models:
Policy Optimization for Diffusion Models:
Policy Optimization for Language Models:
Reinforcement Learning for Visual Generation: