Zhixiong Xu

About

Zhixiong Xu is a leading researcher at the forefront of deep reinforcement learning (DRL), with a focused mission to overcome its most critical limitation: sample inefficiency and slow task adaptation. Xu’s work is centered on developing sophisticated meta-learning frameworks that enable AI agents to "learn to learn" more effectively. A key contribution is the introduction of hierarchical meta-critic networks, which allow for more efficient policy learning by structuring the learning process across multiple levels of abstraction. Further advancing the field, Xu proposed a novel combination of model-based and gradient-based meta-learning, achieving fast task adaptation in complex environments. To refine the learning signal itself, Xu developed a weighted gradient update mechanism, ensuring that the most informative experiences drive the learning process. With seminal papers from 2019 and 2020 accumulating over 60 citations, Xu’s research has directly addressed the data-hungry nature of DRL, paving the way for more practical and adaptable AI systems in robotics, games, and dialogue systems.

Research Focus

Key Achievements

Top Papers

1. 1
   Fast Task Adaptation Based on the Combination of Model-Based and Gradient-Based Meta Learning

   26 citations · 2020
2. 2
   Learning to Learn: Hierarchical Meta-Critic Networks

   23 citations · 2019
3. 3
   Meta-Learning via Weighted Gradient Update

   12 citations · 2019

Key Collaborators

Contact & Links