标题： virtuously Safe Reinforcement Learning 显示英文标题

标题： Virtuously Safe Reinforcement Learning

摘要： 我们证明，当一个第三方，即攻击者，进入安全可中断强化学习的两方设置（代理和操作员）时，必须在极限情况下遵循最优策略的概率和逃避攻击者造成的危险情况的概率之间做出权衡。 到目前为止，关于安全可中断代理的研究假设代理对其环境有完美的感知（没有攻击者），因此通过明确追求第一个概率为一，隐式地将第二个概率设为零。 我们证明了（1）代理可以同时具备可中断性和抗攻击性，（2）可中断性可以被设计成安全的，即代理本身不会试图避免它。 我们还解决了当代理不完全贪婪时出现的问题，即极限情况下的安全探索问题。 对扰动感知的鲁棒性、极限情况下的安全探索以及安全可中断性是我们在文中称为\emph{虚拟安全强化学习}的三个支柱。 显示英文摘要

摘要： We show that when a third party, the adversary, steps into the two-party setting (agent and operator) of safely interruptible reinforcement learning, a trade-off has to be made between the probability of following the optimal policy in the limit, and the probability of escaping a dangerous situation created by the adversary. So far, the work on safely interruptible agents has assumed a perfect perception of the agent about its environment (no adversary), and therefore implicitly set the second probability to zero, by explicitly seeking a value of one for the first probability. We show that (1) agents can be made both interruptible and adversary-resilient, and (2) the interruptibility can be made safe in the sense that the agent itself will not seek to avoid it. We also solve the problem that arises when the agent does not go completely greedy, i.e. issues with safe exploration in the limit. Resilience to perturbed perception, safe exploration in the limit, and safe interruptibility are the three pillars of what we call \emph{virtuously safe reinforcement learning}.