We study matrix estimation problems arising in reinforcement learning (RL) with low-rank structure. In low-rank bandits, the matrix to be recovered specifies the expected arm rewards, and for low-rank Markov Decision Processes (MDPs), it may for example characterize the transition kernel of the MDP. In both cases, each entry of the matrix carries important information, and we seek estimation methods with low entry-wise error. Importantly, these methods further need to accommodate for inherent correlations in the available data (e.g. for MDPs, the data consists of system trajectories). We investigate the performance of simple spectral-based matrix estimation approaches: we show that they efficiently recover the singular subspaces of the matrix and exhibit nearly-minimal entry-wise error. These new results on low-rank matrix estimation make it possible to devise reinforcement learning algorithms that fully exploit the underlying low-rank structure. We provide two examples of such algorithms: a regret minimization algorithm for low-rank bandit problems, and a best policy identification algorithm for reward-free RL in low-rank MDPs. Both algorithms yield state-of-the-art performance guarantees.

Reinforcement learning agents deployed in the real world often have to cope with partially observable environments. Therefore, most agents employ memory mechanisms to approximate the state of the environment. Recently, there have been impressive success stories in mastering partially observable environments, mostly in the realm of computer games like Dota 2, StarCraft II, or MineCraft. However, existing methods lack interpretability in the sense that it is not comprehensible for humans what the agent stores in its memory. In this regard, we propose a novel memory mechanism that represents past events in human language. Our method uses CLIP to associate visual inputs with language tokens. Then we feed these tokens to a pretrained language model that serves the agent as memory and provides it with a coherent and human-readable representation of the past. We train our memory mechanism on a set of partially observable environments and find that it excels on tasks that require a memory component, while mostly attaining performance on-par with strong baselines on tasks that do not. On a challenging continuous recognition task, where memorizing the past is crucial, our memory mechanism converges two orders of magnitude faster than prior methods. Since our memory mechanism is human-readable, we can peek at an agent's memory and check whether crucial pieces of information have been stored. This significantly enhances troubleshooting and paves the way toward more interpretable agents.

While numerous works have focused on devising efficient algorithms for reinforcement learning (RL) with uniformly bounded rewards, it remains an open question whether sample or time-efficient algorithms for RL with large state-action space exist when the rewards are \emph{heavy-tailed}, i.e., with only finite $(1+\epsilon)$-th moments for some $\epsilon\in(0,1]$. In this work, we address the challenge of such rewards in RL with linear function approximation. We first design an algorithm, \textsc{Heavy-OFUL}, for heavy-tailed linear bandits, achieving an \emph{instance-dependent} $T$-round regret of $\tilde{O}\big(d T^{\frac{1-\epsilon}{2(1+\epsilon)}} \sqrt{\sum_{t=1}^T \nu_t^2} + d T^{\frac{1-\epsilon}{2(1+\epsilon)}}\big)$, the \emph{first} of this kind. Here, $d$ is the feature dimension, and $\nu_t^{1+\epsilon}$ is the $(1+\epsilon)$-th central moment of the reward at the $t$-th round. We further show the above bound is minimax optimal when applied to the worst-case instances in stochastic and deterministic linear bandits. We then extend this algorithm to the RL settings with linear function approximation. Our algorithm, termed as \textsc{Heavy-LSVI-UCB}, achieves the \emph{first} computationally efficient \emph{instance-dependent} $K$-episode regret of $\tilde{O}(d \sqrt{H \mathcal{U}^*} K^\frac{1}{1+\epsilon} + d \sqrt{H \mathcal{V}^* K})$. Here, $H$ is length of the episode, and $\mathcal{U}^*, \mathcal{V}^*$ are instance-dependent quantities scaling with the central moment of reward and value functions, respectively. We also provide a matching minimax lower bound $\Omega(d H K^{\frac{1}{1+\epsilon}} + d \sqrt{H^3 K})$ to demonstrate the optimality of our algorithm in the worst case. Our result is achieved via a novel robust self-normalized concentration inequality that may be of independent interest in handling heavy-tailed noise in general online regression problems.

We initiate the mathematical study of replicability as an algorithmic property in the context of reinforcement learning (RL). We focus on the fundamental setting of discounted tabular MDPs with access to a generative model. Inspired by Impagliazzo et al. [2022], we say that an RL algorithm is replicable if, with high probability, it outputs the exact same policy after two executions on i.i.d. samples drawn from the generator when its internal randomness is the same. We first provide an efficient $\rho$-replicable algorithm for $(\varepsilon, \delta)$-optimal policy estimation with sample and time complexity $\widetilde O\left(\frac{N^3\cdot\log(1/\delta)}{(1-\gamma)^5\cdot\varepsilon^2\cdot\rho^2}\right)$, where $N$ is the number of state-action pairs. Next, for the subclass of deterministic algorithms, we provide a lower bound of order $\Omega\left(\frac{N^3}{(1-\gamma)^3\cdot\varepsilon^2\cdot\rho^2}\right)$. Then, we study a relaxed version of replicability proposed by Kalavasis et al. [2023] called TV indistinguishability. We design a computationally efficient TV indistinguishable algorithm for policy estimation whose sample complexity is $\widetilde O\left(\frac{N^2\cdot\log(1/\delta)}{(1-\gamma)^5\cdot\varepsilon^2\cdot\rho^2}\right)$. At the cost of $\exp(N)$ running time, we transform these TV indistinguishable algorithms to $\rho$-replicable ones without increasing their sample complexity. Finally, we introduce the notion of approximate-replicability where we only require that two outputted policies are close under an appropriate statistical divergence (e.g., Renyi) and show an improved sample complexity of $\widetilde O\left(\frac{N\cdot\log(1/\delta)}{(1-\gamma)^5\cdot\varepsilon^2\cdot\rho^2}\right)$.

Most reinforcement learning methods rely heavily on dense, well-normalized environment rewards. DreamerV3 recently introduced a model-based method with a number of tricks that mitigate these limitations, achieving state-of-the-art on a wide range of benchmarks with a single set of hyperparameters. This result sparked discussion about the generality of the tricks, since they appear to be applicable to other reinforcement learning algorithms. Our work applies DreamerV3's tricks to PPO and is the first such empirical study outside of the original work. Surprisingly, we find that the tricks presented do not transfer as general improvements to PPO. We use a high quality PPO reference implementation and present extensive ablation studies totaling over 10,000 A100 hours on the Arcade Learning Environment and the DeepMind Control Suite. Though our experiments demonstrate that these tricks do not generally outperform PPO, we identify cases where they succeed and offer insight into the relationship between the implementation tricks. In particular, PPO with these tricks performs comparably to PPO on Atari games with reward clipping and significantly outperforms PPO without reward clipping.

Recently, data augmentation (DA) has emerged as a method for leveraging domain knowledge to inexpensively generate additional data in reinforcement learning (RL) tasks, often yielding substantial improvements in data efficiency. While prior work has demonstrated the utility of incorporating augmented data directly into model-free RL updates, it is not well-understood when a particular DA strategy will improve data efficiency. In this paper, we seek to identify general aspects of DA responsible for observed learning improvements. Our study focuses on sparse-reward tasks with dynamics-invariant data augmentation functions, serving as an initial step towards a more general understanding of DA and its integration into RL training. Experimentally, we isolate three relevant aspects of DA: state-action coverage, reward density, and the number of augmented transitions generated per update (the augmented replay ratio). From our experiments, we draw two conclusions: (1) increasing state-action coverage often has a much greater impact on data efficiency than increasing reward density, and (2) decreasing the augmented replay ratio substantially improves data efficiency. In fact, certain tasks in our empirical study are solvable only when the replay ratio is sufficiently low.

Deep reinforcement learning (RL) can enable robots to autonomously acquire complex behaviors, such as legged locomotion. However, RL in the real world is complicated by constraints on efficiency, safety, and overall training stability, which limits its practical applicability. We present APRL, a policy regularization framework that modulates the robot's exploration over the course of training, striking a balance between flexible improvement potential and focused, efficient exploration. APRL enables a quadrupedal robot to efficiently learn to walk entirely in the real world within minutes and continue to improve with more training where prior work saturates in performance. We demonstrate that continued training with APRL results in a policy that is substantially more capable of navigating challenging situations and is able to adapt to changes in dynamics with continued training.

Deep reinforcement learning methods exhibit impressive performance on a range of tasks but still struggle on hard exploration tasks in large environments with sparse rewards. To address this, intrinsic rewards can be generated using forward model prediction errors that decrease as the environment becomes known, and incentivize an agent to explore novel states. While prediction-based intrinsic rewards can help agents solve hard exploration tasks, they can suffer from catastrophic forgetting and actually increase at visited states. We first examine the conditions and causes of catastrophic forgetting in grid world environments. We then propose a new method FARCuriosity, inspired by how humans and animals learn. The method depends on fragmentation and recall: an agent fragments an environment based on surprisal, and uses different local curiosity modules (prediction-based intrinsic reward functions) for each fragment so that modules are not trained on the entire environment. At each fragmentation event, the agent stores the current module in long-term memory (LTM) and either initializes a new module or recalls a previously stored module based on its match with the current state. With fragmentation and recall, FARCuriosity achieves less forgetting and better overall performance in games with varied and heterogeneous environments in the Atari benchmark suite of tasks. Thus, this work highlights the problem of catastrophic forgetting in prediction-based curiosity methods and proposes a solution.

The efficient deployment and fine-tuning of foundation models are pivotal in contemporary artificial intelligence. In this study, we present a groundbreaking paradigm integrating Mobile Edge Computing (MEC) with foundation models, specifically designed to enhance local task performance on user equipment (UE). Central to our approach is the innovative Emulator-Adapter architecture, segmenting the foundation model into two cohesive modules. This design not only conserves computational resources but also ensures adaptability and fine-tuning efficiency for downstream tasks. Additionally, we introduce an advanced resource allocation mechanism that is fine-tuned to the needs of the Emulator-Adapter structure in decentralized settings. To address the challenges presented by this system, we employ a hybrid multi-agent Deep Reinforcement Learning (DRL) strategy, adept at handling mixed discrete-continuous action spaces, ensuring dynamic and optimal resource allocations. Our comprehensive simulations and validations underscore the practical viability of our approach, demonstrating its robustness, efficiency, and scalability. Collectively, this work offers a fresh perspective on deploying foundation models and balancing computational efficiency with task proficiency.

The emergence of foundation models, including language and vision models, has reshaped AI's landscape, offering capabilities across various applications. Deploying and fine-tuning these large models, like GPT-3 and BERT, presents challenges, especially in the current foundation model era. We introduce Emulator-Assisted Tuning (EAT) combined with Parameter-Efficient Fine-Tuning (PEFT) to form Parameter-Efficient Emulator-Assisted Tuning (PEAT). Further, we expand this into federated learning as Federated PEAT (FedPEAT). FedPEAT uses adapters, emulators, and PEFT for federated model tuning, enhancing model privacy and memory efficiency. Adapters adjust pre-trained models, while emulators give a compact representation of original models, addressing both privacy and efficiency. Adaptable to various neural networks, our approach also uses deep reinforcement learning for hyper-parameter optimization. We tested FedPEAT in a unique scenario with a server participating in collaborative federated tuning, showcasing its potential in tackling foundation model challenges.