In this work we introduce reinforcement learning techniques for solving lexicographic multi-objective problems. These are problems that involve multiple reward signals, and where the goal is to learn a policy that maximises the first reward signal, and subject to this constraint also maximises the second reward signal, and so on. We present a family of both action-value and policy gradient algorithms that can be used to solve such problems, and prove that they converge to policies that are lexicographically optimal. We evaluate the scalability and performance of these algorithms empirically, demonstrating their practical applicability. As a more specific application, we show how our algorithms can be used to impose safety constraints on the behaviour of an agent, and compare their performance in this context with that of other constrained reinforcement learning algorithms.

The aim of Inverse Reinforcement Learning (IRL) is to infer a reward function $R$ from a policy $\pi$. To do this, we need a model of how $\pi$ relates to $R$. In the current literature, the most common models are optimality, Boltzmann rationality, and causal entropy maximisation. One of the primary motivations behind IRL is to infer human preferences from human behaviour. However, the true relationship between human preferences and human behaviour is much more complex than any of the models currently used in IRL. This means that they are misspecified, which raises the worry that they might lead to unsound inferences if applied to real-world data. In this paper, we provide a mathematical analysis of how robust different IRL models are to misspecification, and answer precisely how the demonstrator policy may differ from each of the standard models before that model leads to faulty inferences about the reward function $R$. We also introduce a framework for reasoning about misspecification in IRL, together with formal tools that can be used to easily derive the misspecification robustness of new IRL models.

我们提供了奖励黑客的第一个正式定义，即优化不完美的代理奖励功能的现象，$ \ Mathcal {\ tilde {r}} $，根据真实的奖励功能，$ \ MATHCAL {R} $导致性能差。 。我们说，如果增加预期的代理回报率永远无法减少预期的真实回报，则代理是不可接受的。直觉上，可以通过从奖励功能（使其“较窄”）中留出一些术语或忽略大致等效的结果之间的细粒度区分来创建一个不可接受的代理，但是我们表明情况通常不是这样。一个关键的见解是，奖励的线性性（在州行动访问计数中）使得无法实现的状况非常强烈。特别是，对于所有随机策略的集合，只有在其中一个是恒定的，只有两个奖励函数才能是不可接受的。因此，我们将注意力转移到确定性的政策和有限的随机政策集中，在这些策略中，始终存在非平凡的不可动摇的对，并为简化的存在建立必要和充分的条件，这是一个重要的不被限制的特殊情况。我们的结果揭示了使用奖励函数指定狭窄任务和对齐人类价值的AI系统之间的紧张关系。

我们分析了学习型号（如神经网络）本身是优化器时发生的学习优化的类型 - 我们将作为MESA优化的情况，我们在本文中介绍的新闻。我们认为，MESA优化的可能性为先进机器学习系统的安全和透明度提出了两个重要问题。首先，在什么情况下学习模型是优化的，包括当他们不应该？其次，当学习模型是优化器时，它的目标是什么 - 它将如何与损失函数不同，它训练的损失 - 并且如何对齐？在本文中，我们对这两个主要问题进行了深入的分析，并提供了未来研究的主题概述。

我们开发一种方法来构造来自表示基本上非线性（或不可连锁的）动态系统的数据集构成低维预测模型，其中具有由有限许多频率的外部强制进行外部矫正的双曲线线性部分。我们的数据驱动，稀疏，非线性模型获得为低维，吸引动力系统的光谱子纤维（SSM）的降低的动态的延长正常形式。我们说明了数据驱动的SSM降低了高维数值数据集的功率和涉及梁振荡，涡旋脱落和水箱中的晃动的实验测量。我们发现，在未加工的数据上培训的SSM减少也在额外的外部强制下准确预测非线性响应。