Hao Zhou

BFP-based NPUs, despite their efficiency, exhibit surprising vulnerabilities to hardware faults that can be mitigated with a novel microarchitecture incurring minimal overhead.

Fine-grained partitioning and targeted safeguards can provide robust NPU reliability with minimal performance overhead, challenging the assumption that redundancy is the only path forward.

Predicting when drivers will engage or disengage driving automation requires more than just video—CAN bus data and route context are key, and handover/takeover events exhibit distinct temporal dependencies.

Forget brittle, hand-coded robot assembly routines: ATG-MoE learns complex, multi-skill manipulation directly from visual and language inputs, achieving impressive success rates in both simulation and real-world industrial tasks.

Incomplete trajectory data got you down? This plug-and-play framework progressively aligns features from incomplete observations with complete ones, boosting prediction accuracy in autonomous driving scenarios.

Agentic RL can now beat proprietary LLMs and torch.compile in the challenging domain of CUDA kernel generation, achieving up to 40% speedups on hard tasks.

LLMs promise to revolutionize UAVs by enabling advanced environmental understanding, swarm coordination, and high-level task reasoning, paving the way for more adaptive and context-aware aerial operations.

Telecom LLMs get a 14% accuracy boost by fusing knowledge graphs with retrieval-augmented generation, slashing hallucinations and improving reliability.

Achieve up to 12% improvement in Rouge-LSum for multimodal tasks in edge-cloud settings by jointly training heterogeneous edge models with a server model, even with varying modality availability.