Cheng Wang

Autonomous model post-training can outperform human-crafted instruct baselines on hard coding benchmarks while completely eliminating reward hacking once experimental exploration is governed by a sandboxed OS and a dynamic peer-review DAG.

Achieving superior instruction accuracy and edit fidelity, GMO-E$^2$DIT redefines how we approach multi-operation image editing in e-commerce.

Stagnant neurons in MARL can cripple learning, but KNIFE revitalizes them, leading to substantial performance gains in complex environments.

MotionMAR achieves unprecedented accuracy in human motion reconstruction from sparse data by effectively separating global trajectories from fine details.

CIM accelerators can outperform GPUs in space by maintaining efficient heat distribution, crucial for high-performance AI workloads under thermal constraints.

Tether-aware collision avoidance can significantly enhance the operational safety of USV-HROV systems in complex marine environments.

Forget expensive expert annotation – this framework synthesizes realistic task-oriented dialogues at scale by mining publicly available streaming media.

Fragmented privacy patches are insufficient for Embodied AI: a unified, lifecycle-level approach is needed to prevent systemic privacy leakage in real-world deployments.