Tian Li

This paper proposes two post-processing techniques, random selection and linear combination, to construct a model that satisfies any desired differential privacy level without retraining, given a set of existing models.

VITAL introduces a novel latent-space reasoning framework for medical MLLMs, utilizing visual-semantic dual supervision to enhance reasoning capabilities and provide crucial interpretability without sacrificing efficiency.

The paper introduces MTAVG-Bench 2.0, a new benchmark designed to diagnose high-level failure modes of cinematic expressiveness in multi-talker audio-video generation, showing that even advanced models struggle with complex scene-level failures.

DELOS is a novel contrastive-learning framework that efficiently and sensitively detects shallow, intermediate-to-long-period exoplanet transits in Kepler photometry, significantly outperforming traditional methods like BLS and TLS in low Signal-to-Noise Ratio regimes.

ReSkill is an RL-in-the-loop framework that reconciles skill creation and policy optimization by automatically creating, testing, and refining modular skills alongside the agent's policy learning, leading to superior generalization.

This paper introduces Arachne, a framework for efficient Text-to-Video model training at scale, reducing iteration time by up to 65% over leading frameworks.

The paper introduces MedPMC, a framework that transforms permissively licensed literature into high-fidelity infrastructure for medical multimodal models, resulting in improved performance on various benchmarks.

This paper proposes Yi, a system for efficient and effective in-place updates in large-scale vector indexing, achieving higher update and search throughput than state-of-the-art methods.

This paper analyzes large-scale human-LLM conversations to identify learning behaviors and their associated factors.

A neural network called DMSNet is proposed for multi-target sensing in multi-band integrated sensing and communication systems, improving accuracy, reducing errors, and decreasing runtime.

This paper introduces Skill Self-Play (Skill-SP), a co-evolutionary framework for LLM training that bridges the gap between structured verification and open-ended exploration.