Sirui Han

This paper proposes HDR (Hierarchical Denoising for Visual Reasoning), a framework that integrates hierarchical latents into causal video generation for multi-step reasoning, enabling coarse-to-fine reasoning and reducing temporal attention costs.

This paper organizes embodied data sources for multimodal foundation models into a pyramid, focusing on real-robot, UMI-style, egocentric and exocentric, simulation, and general vision-language data.

The paper introduces SkillProx, a framework for improving LLM agents by coupling closed-loop diagnostic evolution with utility-aware proximal refinement.

This paper evaluates the use of rendered code as operational context for agentic coding, showing consistent reduction in prompt-token cost but limited increase in linear savings and end-to-end repair accuracy.