Yingxiao Xiang

About

Yingxiao Xiang is a leading researcher at the intersection of artificial intelligence, deep reinforcement learning (DRL), and adversarial machine learning. Their work focuses on understanding and fortifying the security of autonomous systems, particularly in pathfinding and robot control. Xiang’s pioneering contributions include the development of adversarial examples that attack Q-learning and DQN pathfinding algorithms, as demonstrated in their highly cited 2018 papers, which have garnered 25 and 23 citations respectively. These studies revealed critical vulnerabilities in DRL-based navigation, showing how subtle perturbations can derail autonomous agents. Building on this, Xiang introduced a PCA-based model to predict such attacks and, more recently, a curricular robust reinforcement learning framework using GAN-based perturbations (2022, 12 citations). This innovative approach schedules task sequences to enhance robustness, addressing a key challenge in deploying RL in real-world distributed systems like cooperative robots. Xiang’s work has been instrumental in advancing the safety and reliability of AI, making them a notable figure in adversarial robustness research.

Research Focus

Key Achievements

Top Papers

1. 1
   Adversarial Examples Construction Towards White-Box Q Table Variation in DQN Pathfinding Training

   25 citations · 2018
2. 2
   A PCA-Based Model to Predict Adversarial Examples on Q-Learning of Path Finding

   23 citations · 2018
3. 3
   Curricular Robust Reinforcement Learning via GAN-Based Perturbation Through Continuously Scheduled Task Sequence

   12 citations · 2022

Key Collaborators

Contact & Links