A Method for Evaluating Hyperparameter Sensitivity in Reinforcement Learning

The performance of modern reinforcement learning algorithms critically relies on tuning ever-increasing numbers of hyperparameters. Often, small changes in a hyperparameter can lead to drastic changes in performance, and different environments require very different hyperparameter settings to achieve state-of-the-art performance reported in the literature. We currently lack a scalable and widely accepted approach to characterizing these complex interactions. This work proposes a new empirical methodology for studying, comparing, and quantifying the sensitivity of an algorithm's performance to hyperparameter tuning for a given set of environments. We then demonstrate the utility of this methodology by assessing the hyperparameter sensitivity of several commonly used normalization variants of PPO. The results suggest that several algorithmic performance improvements may, in fact, be a result of an increased reliance on hyperparameter tuning.

Real-Time Recurrent Learning using Trace Units in Reinforcement Learning

Recurrent Neural Networks (RNNs) are used to learn representations in partially observable environments. For agents that learn online and continually interact with the environment, it is desirable to train RNNs with real-time recurrent learning (RTRL); unfortunately, RTRL is prohibitively expensive for standard RNNs. A promising direction is to use linear recurrent architectures (LRUs), where dense recurrent weights are replaced with a complex-valued diagonal, making RTRL efficient. In this work, we build on these insights to provide a lightweight but effective approach for training RNNs in online RL. We introduce Recurrent Trace Units (RTUs), a small modification on LRUs that we nonetheless find to have significant performance benefits over LRUs when trained with RTRL. We find RTUs significantly outperform other recurrent architectures across several partially observable environments while using significantly less computation.

Meta-Gradient Search Control: A Method for Improving the Efficiency of Dyna-style Planning

We study how a Reinforcement Learning (RL) system can remain sample-efficient when learning from an imperfect model of the environment. This is particularly challenging when the learning system is resource-constrained and in continual settings, where the environment dynamics change. To address these challenges, our paper introduces an online, meta-gradient algorithm that tunes a probability with which states are queried during Dyna-style planning. Our study compares the aggregate, empirical performance of this meta-gradient method to baselines that employ conventional sampling strategies. Results indicate that our method improves efficiency of the planning process, which, as a consequence, improves the sample-efficiency of the overall learning process. On the whole, we observe that our meta-learned solutions avoid several pathologies of conventional planning approaches, such as sampling inaccurate transitions and those that stall credit assignment. We believe these findings could prove useful, in future work, for designing model-based RL systems at scale.

Beyond Optimism: Exploration With Partially Observable Rewards

Exploration in reinforcement learning (RL) remains an open challenge. RL algorithms rely on observing rewards to train the agent, and if informative rewards are sparse the agent learns slowly or may not learn at all. To improve exploration and reward discovery, popular algorithms rely on optimism. But what if sometimes rewards are unobservable, e.g., situations of partial monitoring in bandits and the recent formalism of monitored Markov decision process? In this case, optimism can lead to suboptimal behavior that does not explore further to collapse uncertainty. With this paper, we present a novel exploration strategy that overcomes the limitations of existing methods and guarantees convergence to an optimal policy even when rewards are not always observable. We further propose a collection of tabular environments for benchmarking exploration in RL (with and without unobservable rewards) and show that our method outperforms existing ones.

Monitored Markov Decision Processes

In reinforcement learning (RL), an agent learns to perform a task by interacting with an environment and receiving feedback (a numerical reward) for its actions. However, the assumption that rewards are always observable is often not applicable in real-world problems. For example, the agent may need to ask a human to supervise its actions or activate a monitoring system to receive feedback. There may even be a period of time before rewards become observable, or a period of time after which rewards are no longer given. In other words, there are cases where the environment generates rewards in response to the agent's actions but the agent cannot observe them. In this paper, we formalize a novel but general RL framework - Monitored MDPs - where the agent cannot always observe rewards. We discuss the theoretical and practical consequences of this setting, show challenges raised even in toy environments, and propose algorithms to begin to tackle this novel setting. This paper introduces a powerful new formalism that encompasses both new and existing problems and lays the foundation for future research.

Assessing the Interpretability of Programmatic Policies with Large Language Models

Although the synthesis of programs encoding policies often carries the promise of interpretability, systematic evaluations to assess the interpretability of these policies were never performed, likely because of the complexity of such an evaluation. In this paper, we introduce a novel metric that uses large-language models (LLM) to assess the interpretability of programmatic policies. For our metric, an LLM is given both a program and a description of its associated programming language. The LLM then formulates a natural language explanation of the program. This explanation is subsequently fed into a second LLM, which tries to reconstruct the program from the natural language explanation. Our metric measures the behavioral similarity between the reconstructed program and the original. We validate our approach using obfuscated programs that are used to solve classic programming problems. We also assess our metric with programmatic policies synthesized for playing a real-time strategy game, comparing the interpretability scores of programmatic policies synthesized by an existing system to lightly obfuscated versions of the same programs. Our LLM-based interpretability score consistently ranks less interpretable programs lower and more interpretable ones higher. These findings suggest that our metric could serve as a reliable and inexpensive tool for evaluating the interpretability of programmatic policies.

TacticAI: an AI assistant for football tactics

Identifying key patterns of tactics implemented by rival teams, and developing effective responses, lies at the heart of modern football. However, doing so algorithmically remains an open research challenge. To address this unmet need, we propose TacticAI, an AI football tactics assistant developed and evaluated in close collaboration with domain experts from Liverpool FC. We focus on analysing corner kicks, as they offer coaches the most direct opportunities for interventions and improvements. TacticAI incorporates both a predictive and a generative component, allowing the coaches to effectively sample and explore alternative player setups for each corner kick routine and to select those with the highest predicted likelihood of success. We validate TacticAI on a number of relevant benchmark tasks: predicting receivers and shot attempts and recommending player position adjustments. The utility of TacticAI is validated by a qualitative study conducted with football domain experts at Liverpool FC. We show that TacticAI's model suggestions are not only indistinguishable from real tactics, but also favoured over existing tactics 90% of the time, and that TacticAI offers an effective corner kick retrieval system. TacticAI achieves these results despite the limited availability of gold-standard data, achieving data efficiency through geometric deep learning.

Proper Laplacian Representation Learning

The ability to learn good representations of states is essential for solving large reinforcement learning problems, where exploration, generalization, and transfer are particularly challenging. The Laplacian representation is a promising approach to address these problems by inducing intrinsic rewards for temporally-extended action discovery and reward shaping, and informative state encoding. To obtain the Laplacian representation one needs to compute the eigensystem of the graph Laplacian, which is often approximated through optimization objectives compatible with deep learning approaches. These approximations, however, depend on hyperparameters that are impossible to tune efficiently, converge to arbitrary rotations of the desired eigenvectors, and are unable to accurately recover the corresponding eigenvalues. In this paper we introduce a theoretically sound objective and corresponding optimization algorithm for approximating the Laplacian representation. Our approach naturally recovers both the true eigenvectors and eigenvalues while eliminating the hyperparameter dependence of previous approximations. We provide theoretical guarantees for our method and we show that those results translate empirically into robust learning across multiple environments.

Targeted Search Control in AlphaZero for Effective Policy Improvement

AlphaZero is a self-play reinforcement learning algorithm that achieves superhuman play in chess, shogi, and Go via policy iteration. To be an effective policy improvement operator, AlphaZero's search requires accurate value estimates for the states appearing in its search tree. AlphaZero trains upon self-play matches beginning from the initial state of a game and only samples actions over the first few moves, limiting its exploration of states deeper in the game tree. We introduce Go-Exploit, a novel search control strategy for AlphaZero. Go-Exploit samples the start state of its self-play trajectories from an archive of states of interest. Beginning self-play trajectories from varied starting states enables Go-Exploit to more effectively explore the game tree and to learn a value function that generalizes better. Producing shorter self-play trajectories allows Go-Exploit to train upon more independent value targets, improving value training. Finally, the exploration inherent in Go-Exploit reduces its need for exploratory actions, enabling it to train under more exploitative policies. In the games of Connect Four and 9x9 Go, we show that Go-Exploit learns with a greater sample efficiency than standard AlphaZero, resulting in stronger performance against reference opponents and in head-to-head play. We also compare Go-Exploit to KataGo, a more sample efficient reimplementation of AlphaZero, and demonstrate that Go-Exploit has a more effective search control strategy. Furthermore, Go-Exploit's sample efficiency improves when KataGo's other innovations are incorporated.

Settling the Reward Hypothesis

The reward hypothesis posits that, "all of what we mean by goals and purposes can be well thought of as maximization of the expected value of the cumulative sum of a received scalar signal (reward)." We aim to fully settle this hypothesis. This will not conclude with a simple affirmation or refutation, but rather specify completely the implicit requirements on goals and purposes under which the hypothesis holds.