Rui Li

A new scheduling framework cuts LLM latency by over 10% while enhancing fairness, challenging the status quo of rigid scheduling policies.

Achieving superior safety alignment with LLMs using only 100 harmful samples, SafeSteer drastically cuts alignment costs while maintaining model performance.

Single-view RGB input can revolutionize how robots perceive and manipulate transparent objects, achieving reliable grasping without complex depth sensing.

Stop hand-engineering your multi-agent LLM systems: UnityMAS-O lets you train them end-to-end with RL, unlocking surprisingly large gains, especially for smaller models.

Ditch the high-degree polynomials: decision-aware quadratic ReLU replacements can slash homomorphic encryption inference time by up to 4x without sacrificing accuracy.

Stop blasting your diffusion models with a single, static reward signal: fine-grained credit assignment across denoising steps and objectives unlocks better image and video generation.

Forget imitation: reward-aware trajectory shaping lets few-step generative models outperform their multi-step teachers.

Stop hard-coding reasoning strategies for your LLM agent: a learned router that dynamically picks the best paradigm for each task boosts performance by up to 5.5%, beating even the best fixed strategy.