Hui Wang

Achieving unprecedented geometric fidelity in 3D reconstructions of pelvic organs could revolutionize patient-specific modeling and analysis in medical applications.

Achieving 82% sparsity with minimal accuracy loss, this method redefines efficiency in video generation for Diffusion Transformers.

Achieving six times the inference throughput of current LLMs while maintaining accuracy, Nemotron 3 Ultra redefines performance benchmarks for agentic reasoning tasks.

Achieving state-of-the-art low-bitrate performance with a 4x reduction in codebook size, self-guidance revolutionizes neural audio coding without sacrificing fidelity.

Missing modalities don’t have to mean missing insights—LWR enables robust multimodal predictions by leveraging only the available data.

Query-adaptive detection of active modalities boosts retrieval accuracy by 11.3% over fixed fusion methods in real-world video archives.

Current knee MRI benchmarks miss the holistic clinical picture; MeniOmni fills this gap with multimodal data and clinically relevant evaluation, revealing that patient context significantly improves diagnostic accuracy.

Training for speech editing with reinforcement learning not only enhances editing quality but also unexpectedly boosts zero-shot TTS performance.

Sub-linear attention is now possible without sacrificing complete long-range dependency retention, thanks to learnable summary tokens that compress context.