Qian Wang

Multi-agent LLM systems for idea generation can backfire, with smarter models and more communication leading to *less* diverse ideas due to structural coupling.

Tiny VLMs can punch far above their weight: a 0.5B parameter model, guided by Switch-KD, closes the gap with its 3B teacher by 3.6 points on multimodal benchmarks.

Robot manipulation models trained on mostly VR data can perform as well as those trained on real-world data, but at 1/20th the cost.

Touchscreen authentication can be dramatically improved by jointly modeling physiological contact structures and behavioral motion dynamics, achieving near-perfect accuracy and strong attack resilience.

Generate high-fidelity 3D environments for robot simulation in seconds using only a single panoramic image.