Methods, systems, and computer program products for securing deep reinforcement learning agents

Methods, systems, and computer program products are provided for preventing unauthorized use of deep reinforcement learning agents. The DRL agent is trained to act as intended only when one or more desired secret operation keys are observed. In some embodiments, the DRL agent is also trained to operate at a reduced capacity when the one or more desired secret operation keys are not in use. 提供了用于防止未经授权使用深度强化学习智能体的方法、系统和计算机程序产品。将所述DRL智能体训练为仅在观察到一个或多个所需的秘密操作密钥时才按预期行事。在一些实施例中，当所述一个或多个所需的秘密操作密钥未使用时，还将所述DRL智能体训练为以减少的容量进行操作。