Partner Researcher · Microsoft Research
Xiaodong Liu
Leading AI inference optimization, large language model post-training, reinforcement learning, and safety alignment at Microsoft Research.
This inference optimization work has delivered billion-dollar cost reductions. Built model families including MT-DNN, DeBERTa, UniLM, METRO, PubMedBERT, and Phi that power products used by billions of people and have collectively surpassed 500 million downloads on Hugging Face.
LLM post-training
Reinforcement learning
Safety alignment
Efficient inference
Foundation models
Research
Current work spans AI inference optimization, post-training, reinforcement learning, and safety alignment at Microsoft scale.
Current Focus
- LLM post-training and reinforcement learning
- Safety alignment and adversarial robustness
- Efficient inference and scalable AI systems
- Foundation models and long-context reasoning
Service
- Senior Area Chair
- NeurIPS 2026
- ICML 2026
- NeurIPS 2025
- ...