Dawn Song

About

Dawn Song is a pioneering researcher whose work lies at the intersection of artificial intelligence, cybersecurity, and privacy. Her major contributions focus on exposing critical vulnerabilities in deep reinforcement learning (DRL) systems, demonstrating how these models can be manipulated through adversarial attacks and how they inadvertently leak sensitive information. In her highly cited 2019 paper, *Characterizing Attacks on Deep Reinforcement Learning* (52 citations), she systematically analyzed how small perturbations to observations can fool DRL models, revealing fundamental security flaws in autonomous decision-making systems. Her companion studies, *How You Act Tells a Lot: Privacy-Leaking Attack on Deep Reinforcement Learning* (38 and 14 citations), broke new ground by showing that DRL agents trained on privacy-sensitive data can inadvertently expose confidential information through their actions alone. This work has profound implications for applications ranging from robotics to healthcare AI. Beyond these specific contributions, Song’s research has shaped how the machine learning community thinks about robustness and data protection. She is also known for her broader leadership in AI safety, having founded initiatives that bridge academic research with real-world deployment challenges. Her work has earned her recognition as a visionary in trustworthy AI.

Research Focus

Key Achievements

Top Papers

1. 1
   Characterizing Attacks on Deep Reinforcement Learning

   52 citations · 2019
2. 2
   How You Act Tells a Lot: Privacy-Leaking Attack on Deep Reinforcement Learning

   38 citations · 2019
3. 3
   How You Act Tells a Lot: Privacy-Leakage Attack on Deep Reinforcement Learning

   14 citations · 2019

Key Collaborators

Contact & Links