We study the offline meta-reinforcement learning (OMRL) problem, a paradigm which enables reinforcement learning (RL) algorithms to quickly adapt to unseen tasks without any interactions with the environments, making RL truly practical in many real-world applications. This problem is still not fully understood, for which two major challenges need to be addressed. First, offline RL often suffers from bootstrapping errors of out-of-distribution state-actions which leads to divergence of value functions. Second, meta-RL requires efficient and robust task inference learned jointly with control policy. In this work, we enforce behavior regularization on learned policy as a general approach to offline RL, combined with a deterministic context encoder for efficient task inference. We propose a novel negative-power distance metric on bounded context embedding space, whose gradients propagation is detached from that of the Bellman backup. We provide analysis and insight showing that some simple design choices can yield substantial improvements over recent approaches involving meta-RL and distance metric learning. To the best of our knowledge, our method is the first model-free and end-to-end OMRL algorithm, which is computationally efficient and demonstrated to outperform prior algorithms on several meta-RL benchmarks.

We introduce Adaptive Procedural Task Generation (APT-Gen), an approach to progressively generate a sequence of tasks as curricula to facilitate reinforcement learning in hard-exploration problems. At the heart of our approach, a task generator learns to create tasks from a parameterized task space via a black-box procedural generation module. To enable curriculum learning in the absence of a direct indicator of learning progress, we propose to train the task generator by balancing the agent's performance in the generated tasks and the similarity to the target tasks. Through adversarial training, the task similarity is adaptively estimated by a task discriminator defined on the agent's experiences, allowing the generated tasks to approximate target tasks of unknown parameterization or outside of the predefined task space. Our experiments on grid world and robotic manipulation task domains show that APT-Gen achieves substantially better performance than various existing baselines by generating suitable tasks of rich variations.

Motivated by the episodic version of the classical inventory control problem, we propose a new Q-learning-based algorithm, Elimination-Based Half-Q-Learning (HQL), that enjoys improved efficiency over existing algorithms for a wide variety of problems in the one-sided-feedback setting. We also provide a simpler variant of the algorithm, Full-Q-Learning (FQL), for the full-feedback setting. We establish that HQL incurs $ \tilde{\mathcal{O}}(H^3\sqrt{ T})$ regret and FQL incurs $\tilde{\mathcal{O}}(H^2\sqrt{ T})$ regret, where $H$ is the length of each episode and $T$ is the total length of the horizon. The regret bounds are not affected by the possibly huge state and action space. Our numerical experiments demonstrate the superior efficiency of HQL and FQL, and the potential to combine reinforcement learning with richer feedback models.

Over the last fifty years, a large body of research in microeconomics has introduced many different mechanisms for resource allocation. Despite the wide variety of available options, "simple" mechanisms such as posted price and serial dictatorship are often preferred for practical applications, including housing allocation [Abdulkadiroğlu and Sönmez, 1998], online procurement [Badanidiyuru et al., 2012], or allocation of medical appointments [Klaus and Nichifor, 2019]. There has been considerable interest in formalizing different notions of simplicity. Li [2017] identifies mechanisms that are particularly simple from a strategic perspective, introducing the concept of obviously strategyproof mechanisms; under obviously strategyproof mechanisms, it is obvious that an agent cannot profit by trying to game the system, as even the worst possible final outcome from behaving truthfully is at least as good as the best possible outcome from any other strategy. Pycia and Troyan [2019] introduce the still stronger concept of strongly obviously strategyproof (SOSP) mechanisms, and show that this class can essentially be identified with sequential price mechanisms, where agents are visited in turn and offered a choice from a menu of options (which may or may not include transfers). SOSP mechanisms are ones in which an agent is not even required to consider her future (truthful) actions to understand that the mechanism is obviously strategyproof.

We investigate the discounting mismatch in actor-critic algorithm implementations from a representation learning perspective. Theoretically, actor-critic algorithms usually have discounting for both actor and critic, i.e., there is a $\gamma^t$ term in the actor update for the transition observed at time $t$ in a trajectory and the critic is a discounted value function. Practitioners, however, usually ignore the discounting ($\gamma^t$) for the actor while using a discounted critic. We investigate this mismatch in two scenarios. In the first scenario, we consider optimizing an undiscounted objective $(\gamma = 1)$ where $\gamma^t$ disappears naturally $(1^t = 1)$. We then propose to interpret the discounting in critic in terms of a bias-variance-representation trade-off and provide supporting empirical results. In the second scenario, we consider optimizing a discounted objective ($\gamma < 1$) and propose to interpret the omission of the discounting in the actor update from an auxiliary task perspective and provide supporting empirical results.

The goal of reinforcement learning is designing algorithms to learn the optimal policy through interactions with the unknown dynamic environment. Markov decision processes (MDPs) plays a central role in reinforcement learning due to their ability to describe the time-independent state transition property. In specific, the discounted MDP is one of the standard MDPs in reinforcement learning to describe sequential tasks without interruption or restart. Various reinforcement learning algorithms have been proposed for discounted MDPs. In specific, Azar et al. (2013) proposed an Empirical QVI algorithm which achieves the optimal sample complexity to find the optimal value function. Sidford et al. (2018a) proposed a sublinear randomized value iteration algorithm that achieves a near-optimal sample complexity to find the optimal policy, and Sidford et al. (2018b) further improved it to reach the optimal sample complexity.

Despite numerous attempts sought to provide empirical evidence of adversarial regularization outperforming sole supervision, the theoretical understanding of such phenomena remains elusive. In this study, we aim to resolve whether adversarial regularization indeed performs better than sole supervision at a fundamental level. To bring this insight into fruition, we study vanishing gradient issue, asymptotic iteration complexity, gradient flow and provable convergence in the context of sole supervision and adversarial regularization. The key ingredient is a theoretical justification supported by empirical evidence of adversarial acceleration in gradient descent. In addition, motivated by a recently introduced unit-wise capacity based generalization bound, we analyze the generalization error in adversarial framework. Guided by our observation, we cast doubts on the ability of this measure to explain generalization. We therefore leave as open questions to explore new measures that can explain generalization behavior in adversarial learning. Furthermore, we observe an intriguing phenomenon in the neural embedded vector space while contrasting adversarial learning with sole supervision.

Action advising is a knowledge exchange mechanism between peers, namely student and teacher, that can help tackle exploration and sample inefficiency problems in deep reinforcement learning. Due to the practical limitations in peer-to-peer communication and the negative implications of over-advising, the peer responsible for initiating these interactions needs to do so only when it's most adequate to exchange advice. Most recently, student-initiated techniques that utilise state novelty and uncertainty estimations have obtained promising results. However, these estimations have several weaknesses, such as having no information regarding the characteristics of convergence and being subject to delays that occur in the presence of experience replay dynamics. We propose a student-initiated action advising algorithm that alleviates these shortcomings. Specifically, we employ Random Network Distillation (RND) to measure the novelty of an advice, for the student to determine whether to proceed with the request; furthermore, we perform RND updates only for the advised states to ensure that the student's convergence will not prevent it from utilising the teacher's knowledge at any stage of learning. Experiments in GridWorld and simplified versions of five Atari games show that our approach can perform on par with the state-of-the-art and demonstrate significant advantages in the scenarios where the existing methods are prone to fail.

We seek to create agents that both act and communicate with other agents in pursuit of a goal. Towards this end, we extend LIGHT (Urbanek et al. 2019)---a large-scale crowd-sourced fantasy text-game---with a dataset of quests. These contain natural language motivations paired with in-game goals and human demonstrations; completing a quest might require dialogue or actions (or both). We introduce a reinforcement learning system that (1) incorporates large-scale language modeling-based and commonsense reasoning-based pre-training to imbue the agent with relevant priors; and (2) leverages a factorized action space of action commands and dialogue, balancing between the two. We conduct zero-shot evaluations using held-out human expert demonstrations, showing that our agents are able to act consistently and talk naturally with respect to their motivations.

Social learning is a key component of human and animal intelligence. By taking cues from the behavior of experts in their environment, social learners can acquire sophisticated behavior and rapidly adapt to new circumstances. This paper investigates whether independent reinforcement learning (RL) agents in a multi-agent environment can use social learning to improve their performance using cues from other agents. We find that in most circumstances, vanilla model-free RL agents do not use social learning, even in environments in which individual exploration is expensive. We analyze the reasons for this deficiency, and show that by introducing a model-based auxiliary loss we are able to train agents to lever-age cues from experts to solve hard exploration tasks. The generalized social learning policy learned by these agents allows them to not only outperform the experts with which they trained, but also achieve better zero-shot transfer performance than solo learners when deployed to novel environments with experts. In contrast, agents that have not learned to rely on social learning generalize poorly and do not succeed in the transfer task. Further,we find that by mixing multi-agent and solo training, we can obtain agents that use social learning to out-perform agents trained alone, even when experts are not avail-able. This demonstrates that social learning has helped improve agents' representation of the task itself. Our results indicate that social learning can enable RL agents to not only improve performance on the task at hand, but improve generalization to novel environments.