Xiaoyu Li

Process evaluations reveal hidden failures in LLM reasoning, showing that lucky successes can mask critical deficiencies in agent performance.

CRPO effectively mitigates exposure bias in self-distillation, leading to superior performance in complex reasoning tasks.

A watermark can not only identify machine-generated text but also attribute it to specific users and extract hidden information, all governed by a precise information-theoretic framework.

Watermarking LLM-agent trajectories just got a major upgrade—TRACE achieves near-perfect detection without sacrificing performance, even under aggressive adversarial conditions.

OPERA's intrinsic reward mechanism enables LLMs to achieve new heights in open-ended reasoning, outperforming proprietary models without the pitfalls of biased supervision.

BoxCtrl's innovative use of RGB 3D bounding boxes enables unprecedented precision in geometric image editing, outperforming traditional methods.

The traditional complexity of leverage-score algorithms is misleading; the real challenge lies in identification, not accuracy, allowing for a dramatic reduction in query complexity.

Reducing end-to-end execution latency in quantum tasks by integrating a tightly-coupled RFSoC interface could revolutionize hybrid classical-quantum computing efficiency.

A novel graph-based dataset pruning method achieves over 40% faster training on ImageNet-1k while maintaining accuracy, challenging the status quo of sample selection.

Long-context LLM rankings dramatically reshuffle when evaluated across a range of context lengths and capabilities, proving that a single headline score is misleading.

Fusing AIS and CCTV data with a cross-modal transformer unlocks more accurate maritime vessel trajectory prediction, even when one data source is sparse or unavailable.

Forget brittle orchestration layers – LLMs can internalize complex reasoning as a learnable "HeavySkill" that rivals external agentic frameworks.