Hang Wu

About Hang Wu

As a Staff Research Scientist at ByteDance, I specialize in applying advanced reinforcement learning (RL) techniques to large-scale systems. My work spans the full ML lifecycle, from pioneering novel RL algorithms for user personalization, large language model (LLM) fine-tuning, to scaling massive recommendation models to billions of parameters on GPU infrastructure. I lead cross-functional teams at ByteDance to architect and deploy these cutting-edge RL solutions for personalization and recommendation, and our work in RL for user personalization have delivered over 20 production launches that have resulted in double-digit growth in user engagement and advertising value. In the LLM domain, I helped designing RL algorithms for RLHF pipelines for customer-facing chatbot applications, pioneered a direct RL solution for RLHF, and also a pipeline for fine-tuning LLM for automatic mathematical theorem proving (ACL’25). My Ph.D. research in Georgia Tech focused on adaptive causal inference algorithms. My dissertation introduced novel methods for improving off-policy learning and meta-learning for causal inference, with applications in healthcare informatics. I developed techniques to reduce dataset bias in medical imaging and designed interpretable deep learning models for identifying causes of death from health records, collaborating closely with the Centers for Disease Control and Prevention (CDC).

Experience

1. Member of Technical Staff

   Perplexity

   Sep 2025 — Present · Palo Alto, CA, US

   reward hacking

Education

• Georgia Institute of Technology

  Doctor of Philosophy (Ph.D.), Machine Learning

  2015 — 2020

• Tsinghua University

  Bachelor of Engineering (B.Eng.), Information Science and Control Theory

  2010 — 2014