Shaoan Xie

Controllable Video Generation with Provable Disentanglement

Provably Learning Task-Relevant World Representation

From Comparison to Composition: Towards Understanding Machine Cognition of Unseen Categories

Attribute-Guided Image Generation with Causally Disentangled Representation

Step-Aware Policy Optimization for Reasoning in Diffusion Large Language Models

Editing by Reconstruction: Background Preservation for Instruction-based Autoregressive Image Editing

Identifiable Interpretation in Generative Models via Causal Minimality

Understanding Compositional Generalization via Hierarchical Concept Models

The Power of Order: Fooling LLMs with Adversarial Table Permutation

Nonparametric Identification of Latent Concepts

Synergy Between Sufficient Changes and Sparse Mixing Procedure for Disentangled Representation Learning

Provably Learning Concepts by Comparison

Learning Vision and Language Concepts for Controllable Image Generation

Learning Identifiable Concepts for Compositional Image Generation