Angel or Demon: Investigating the Plasticity Interventions' Impact on Backdoor Threats in Deep Reinforcement Learning
天使或恶魔:探讨塑性干预对深度强化学习中后门威胁的影响
机构 * Zhejiang University(浙江大学) ; National University of Defense Technology(国防科技大学) ; Xi'an Jiaotong University(西安交通大学)
AI总结 研究通过分析14664个案例,发现仅有一种干预(SAM)加剧了后门威胁,其他干预则缓解了威胁,提出新的稳健后门注入框架和异常损失景观尖锐度作为检测指标。
Comments To appear in the Forty-Third International Conference on Machine Learning (ICML 2026), July 6-11, 2026, Seoul, South Korea