Cheng Huang

Instead of rebuilding the entire recommendation stack from scratch, LIGE-GR generalizes the existing pointwise recommendation system into a listwise generation system, which allows mature recommender systems to benefit from listwise optimization while preserving compatibility with existing models, value functions, and serving infrastructure.

Physics-constraints generative AI can revolutionize tropical cyclone forecasting by dramatically reducing uncertainty and computational demands.

Cross-component filters can dramatically enhance chroma fidelity by exploiting luma-chroma correlations, paving the way for superior video quality.

GlaKG achieves near-perfect classification while providing a transparent reasoning framework that links biomarker evidence to clinical rules, addressing the black-box nature of traditional deep learning models in healthcare.

MVG-KAN achieves superior PM$_{2.5}$ forecasting by leveraging a novel Geo-Wind Graph that captures wind-driven pollutant transport dynamics.

Adaptive reduced-order models can now stay accurate for much longer by remembering their past, thanks to a clever incremental SVD trick.

Fact verification benchmarks may be teaching models the wrong thing: NEI labels are often gamed via dataset-specific shortcuts, meaning models don't actually learn to identify insufficient evidence.

LLM-powered autonomous hacking frameworks often hallucinate attacks rather than execute them, revealing a critical gap between theoretical potential and real-world effectiveness.