Wenqi Zhang

Token-level attribution transforms memory learning, enabling agents to identify and leverage crucial information for better performance in complex tasks.

Image-generation models can outperform text-output VLMs in spatial tasks when answers are expressed directly in pixel space, revealing a critical interface mismatch in current evaluation methods.

VLA-Corrector allows VLA models to adaptively replan actions in real-time, drastically reducing compounding errors in dynamic environments.

LLMs can learn to recognize when they lack sufficient information for reasoning and proactively ask for clarification, leading to more reliable and concise answers.

SFT's instability and reward sparsity can be overcome with a novel Group Fine-Tuning (GFT) framework, leading to better LLM policies.

Offloading memory and computation to a copilot lets a 7B parameter GUI agent outperform larger models on long-horizon tasks, suggesting a path to more efficient and capable GUI automation.

Even frontier models like Claude Sonnet 4.6 stumble when asked to infer user preferences and proactively assist in mobile tasks, achieving less than 50% success despite excelling at explicit task execution.