Kun Li

Closed-loop learning in embodied agents can lead to a staggering 11.1x speedup in inference while achieving record performance on complex tasks.

TGRHuman achieves unprecedented realism in 3D human generation from text, outperforming existing methods in both geometry and texture quality.

Automating rank determination in tensor-based neural networks could revolutionize their application in complex physics simulations.

Crowd4D redefines monocular crowd reconstruction by integrating scene geometry and human dynamics, leading to unprecedented accuracy in complex environments.

Achieving state-of-the-art RRSIS performance with just 2.4% parameter updates challenges the norm of full fine-tuning in multimodal models.

Prospective memory in LLMs is not just harder than retrospective memory; it reveals critical insights into a model's reasoning capacity and attentional robustness.

Unlocking preference representation from noisy latents, PRISM achieves state-of-the-art accuracy while drastically reducing computational costs in video generation.

Achieving 74.419% accuracy, this ensemble approach shows that self-supervised learning can dramatically enhance micro-gesture recognition performance.

ActProbe predicts robot policy failures before they become visually apparent, enhancing both detection accuracy and operational efficiency in real-world tasks.

EgoPriMo enables humanoid robots to generate and forecast complex motions interactively using just egocentric observations and high-level language prompts.

Current AI agents struggle to reliably rediscover scientific knowledge, with top performers averaging only 21.5 out of a possible score, revealing critical gaps in their research capabilities.

AtomWorld achieves the previously impossible: simulating the degradation of reactor pressure vessel steel at the atomistic level across year-and-meter scales.