Difan Zou

Turning Internal Gap into Self-Improvement: Promoting the Generation-Understanding Unification in MLLMs

Learning under Quantization for High-Dimensional Linear Regression

Routing Matters in MoE: Scaling Diffusion Transformers with Explicit Routing Guidance

A Convergence Analysis of Adaptive Optimizers under Floating-point Quantization

Reshaping Reasoning in LLMs: A Theoretical Analysis of RL Training Dynamics through Pattern Selection

Hyper-SET: Designing Transformers via Hyperspherical Energy Minimization

Does Higher Interpretability Imply Better Utility? A Pairwise Analysis on Sparse Autoencoders

On the Complexity Theory of Masked Discrete Diffusion: From $\mathrm{poly}(1/\epsilon)$ to Nearly $\epsilon$-Free

Physics-Informed Distillation of Diffusion Models for PDE-Constrained Generation

On the Collapse Errors Induced by the Deterministic Sampler for Diffusion Models

A Random Matrix Analysis of In-context Memorization for Nonlinear Attention

Theory of Autoregressive Diffusion Model: Inference Complexity and Conditional Dependency Learning

Understanding the Generalization of Stochastic Gradient Adam in Learning Neural Networks

Hierarchical Koopman Diffusion: Fast Generation with Interpretable Diffusion Trajectory

Kernel Regression in Structured Non-IID Settings: Theory and Implications for Denoising Score Learning

On the Robustness of Transformers against Context Hijacking for Linear Classification

Speculative Jacobi-Denoising Decoding for Accelerating Autoregressive Text-to-image Generation

HyPoGen: Optimization-Biased Hypernetworks for Generalizable Policy Generation

How Does Critical Batch Size Scale in Pre-training?

How Does Label Noise Gradient Descent Improve Generalization in the Low SNR Regime?

Can Diffusion Models Learn Hidden Inter-Feature Rules Behind Images?

Masked Autoencoders Are Effective Tokenizers for Diffusion Models

F-Adapter: Frequency-Adaptive Parameter-Efficient Fine-Tuning in Scientific Machine Learning

On the Feature Learning in Diffusion Models

Beyond Surface Structure: A Causal Assessment of LLMs' Comprehension ability

Label Noise Gradient Descent Improves Generalization in the Low SNR Regime

On the Robustness of Transformers against Context Hijacking for Linear Classification

Towards Understanding Fine-Tuning Mechanisms of LLMs via Circuit Analysis

Self-Control of LLM Behaviors by Compressing Suffix Gradient into Prefix Controller

Towards a Theoretical Understanding of Memorization in Diffusion Models