Zhe Li

GameAlpha-2.4K not only revolutionizes RGBA video generation for gaming but also achieves a significant efficiency boost by intelligently bypassing redundant computations.

Replacing standard residual connections with manifold-constrained hyper-connections boosts speaker recognition performance across multiple architectures.

$\omega$-0 enables humanoid robots to seamlessly integrate movement and manipulation, outperforming traditional models by predicting coordinated actions directly from sensory inputs.

Iterative image retrieval can now adapt contextually, achieving a 44.1 R@1 on multi-turn searches—far surpassing previous benchmarks.