Progress in fields of machine learning and adversarial planning has benefited significantly from benchmark domains, from checkers and the classic UCI data sets to Go and Diplomacy. In sequential decision-making, agent evaluation has largely been restricted to few interactions against experts, with the aim to reach some desired level of performance (e.g. beating a human professional player). We propose a benchmark for multiagent learning based on repeated play of the simple game Rock, Paper, Scissors along with a population of forty-three tournament entries, some of which are intentionally sub-optimal. We describe metrics to measure the quality of agents based both on average returns and exploitability. We then show that several RL, online learning, and language model approaches can learn good counter-strategies and generalize well, but ultimately lose to the top-performing bots, creating an opportunity for research in multiagent learning.

Offline reinforcement learning (RL) can in principle synthesize more optimal behavior from a dataset consisting only of suboptimal trials. One way that this can happen is by "stitching" together the best parts of otherwise suboptimal trajectories that overlap on similar states, to create new behaviors where each individual state is in-distribution, but the overall returns are higher. However, in many interesting and complex applications, such as autonomous navigation and dialogue systems, the state is partially observed. Even worse, the state representation is unknown or not easy to define. In such cases, policies and value functions are often conditioned on observation histories instead of states. In these cases, it is not clear if the same kind of "stitching" is feasible at the level of observation histories, since two different trajectories would always have different histories, and thus "similar states" that might lead to effective stitching cannot be leveraged. Theoretically, we show that standard offline RL algorithms conditioned on observation histories suffer from poor sample complexity, in accordance with the above intuition. We then identify sufficient conditions under which offline RL can still be efficient -- intuitively, it needs to learn a compact representation of history comprising only features relevant for action selection. We introduce a bisimulation loss that captures the extent to which this happens, and propose that offline RL can explicitly optimize this loss to aid worst-case sample complexity. Empirically, we show that across a variety of tasks either our proposed loss improves performance, or the value of this loss is already minimized as a consequence of standard offline RL, indicating that it correlates well with good performance.

Ideally, we would place a robot in a real-world environment and leave it there improving on its own by gathering more experience autonomously. However, algorithms for autonomous robotic learning have been challenging to realize in the real world. While this has often been attributed to the challenge of sample complexity, even sample-efficient techniques are hampered by two major challenges - the difficulty of providing well "shaped" rewards, and the difficulty of continual reset-free training. In this work, we describe a system for real-world reinforcement learning that enables agents to show continual improvement by training directly in the real world without requiring painstaking effort to hand-design reward functions or reset mechanisms. Our system leverages occasional non-expert human-in-the-loop feedback from remote users to learn informative distance functions to guide exploration while leveraging a simple self-supervised learning algorithm for goal-directed policy learning. We show that in the absence of resets, it is particularly important to account for the current "reachability" of the exploration policy when deciding which regions of the space to explore. Based on this insight, we instantiate a practical learning system - GEAR, which enables robots to simply be placed in real-world environments and left to train autonomously without interruption. The system streams robot experience to a web interface only requiring occasional asynchronous feedback from remote, crowdsourced, non-expert humans in the form of binary comparative feedback. We evaluate this system on a suite of robotic tasks in simulation and demonstrate its effectiveness at learning behaviors both in simulation and the real world. Project website https://guided-exploration-autonomous-rl.github.io/GEAR/.

Embedding covert streams into a cover channel is a common approach to circumventing Internet censorship, due to censors' inability to examine encrypted information in otherwise permitted protocols (Skype, HTTPS, etc.). However, recent advances in machine learning (ML) enable detecting a range of anti-censorship systems by learning distinct statistical patterns hidden in traffic flows. Therefore, designing obfuscation solutions able to generate traffic that is statistically similar to innocuous network activity, in order to deceive ML-based classifiers at line speed, is difficult. In this paper, we formulate a practical adversarial attack strategy against flow classifiers as a method for circumventing censorship. Specifically, we cast the problem of finding adversarial flows that will be misclassified as a sequence generation task, which we solve with Amoeba, a novel reinforcement learning algorithm that we design. Amoeba works by interacting with censoring classifiers without any knowledge of their model structure, but by crafting packets and observing the classifiers' decisions, in order to guide the sequence generation process. Our experiments using data collected from two popular anti-censorship systems demonstrate that Amoeba can effectively shape adversarial flows that have on average 94% attack success rate against a range of ML algorithms. In addition, we show that these adversarial flows are robust in different network environments and possess transferability across various ML models, meaning that once trained against one, our agent can subvert other censoring classifiers without retraining.

Deep reinforcement learning (RL) has achieved remarkable success in solving complex tasks through its integration with deep neural networks (DNNs) as function approximators. However, the reliance on DNNs has introduced a new challenge called primacy bias, whereby these function approximators tend to prioritize early experiences, leading to overfitting. To mitigate this primacy bias, a reset method has been proposed, which performs periodic resets of a portion or the entirety of a deep RL agent while preserving the replay buffer. However, the use of the reset method can result in performance collapses after executing the reset, which can be detrimental from the perspective of safe RL and regret minimization. In this paper, we propose a new reset-based method that leverages deep ensemble learning to address the limitations of the vanilla reset method and enhance sample efficiency. The proposed method is evaluated through various experiments including those in the domain of safe RL. Numerical results show its effectiveness in high sample efficiency and safety considerations.

This study presents a novel deep reinforcement learning (DRL)-based handover (HO) protocol, called DHO, specifically designed to address the persistent challenge of long propagation delays in low-Earth orbit (LEO) satellite networks' HO procedures. DHO skips the Measurement Report (MR) in the HO procedure by leveraging its predictive capabilities after being trained with a pre-determined LEO satellite orbital pattern. This simplification eliminates the propagation delay incurred during the MR phase, while still providing effective HO decisions. The proposed DHO outperforms the legacy HO protocol across diverse network conditions in terms of access delay, collision rate, and handover success rate, demonstrating the practical applicability of DHO in real-world networks. Furthermore, the study examines the trade-off between access delay and collision rate and also evaluates the training performance and convergence of DHO using various DRL algorithms.

With continuous advances in deep learning, distributed training is becoming common in GPU clusters. Specifically, for emerging workloads with diverse amounts, ratios, and patterns of communication, we observe that network contention can significantly degrade training throughput. However, widely used scheduling policies often face limitations as they are agnostic to network contention between jobs. In this paper, we present a new approach to mitigate network contention in GPU clusters using reinforcement learning. We formulate GPU cluster scheduling as a reinforcement learning problem and opt to learn a network contention-aware scheduling policy that efficiently captures contention sensitivities and dynamically adapts scheduling decisions through continuous evaluation and improvement. We show that compared to widely used scheduling policies, our approach reduces average job completion time by up to 18.2\% and effectively cuts the tail job completion time by up to 20.7\% while allowing a preferable trade-off between average job completion time and resource utilization.

Designing reward functions for efficiently guiding reinforcement learning (RL) agents toward specific behaviors is a complex task. This is challenging since it requires the identification of reward structures that are not sparse and that avoid inadvertently inducing undesirable behaviors. Naively modifying the reward structure to offer denser and more frequent feedback can lead to unintended outcomes and promote behaviors that are not aligned with the designer's intended goal. Although potential-based reward shaping is often suggested as a remedy, we systematically investigate settings where deploying it often significantly impairs performance. To address these issues, we introduce a new framework that uses a bi-level objective to learn \emph{behavior alignment reward functions}. These functions integrate auxiliary rewards reflecting a designer's heuristics and domain knowledge with the environment's primary rewards. Our approach automatically determines the most effective way to blend these types of feedback, thereby enhancing robustness against heuristic reward misspecification. Remarkably, it can also adapt an agent's policy optimization process to mitigate suboptimalities resulting from limitations and biases inherent in the underlying RL algorithms. We evaluate our method's efficacy on a diverse set of tasks, from small-scale experiments to high-dimensional control challenges. We investigate heuristic auxiliary rewards of varying quality -- some of which are beneficial and others detrimental to the learning process. Our results show that our framework offers a robust and principled way to integrate designer-specified heuristics. It not only addresses key shortcomings of existing approaches but also consistently leads to high-performing solutions, even when given misaligned or poorly-specified auxiliary reward functions.

Robust multi-agent reinforcement learning (MARL) necessitates resilience to uncertain or worst-case actions by unknown allies. Existing max-min optimization techniques in robust MARL seek to enhance resilience by training agents against worst-case adversaries, but this becomes intractable as the number of agents grows, leading to exponentially increasing worst-case scenarios. Attempts to simplify this complexity often yield overly pessimistic policies, inadequate robustness across scenarios and high computational demands. Unlike these approaches, humans naturally learn adaptive and resilient behaviors without the necessity of preparing for every conceivable worst-case scenario. Motivated by this, we propose MIR2, which trains policy in routine scenarios and minimize Mutual Information as Robust Regularization. Theoretically, we frame robustness as an inference problem and prove that minimizing mutual information between histories and actions implicitly maximizes a lower bound on robustness under certain assumptions. Further analysis reveals that our proposed approach prevents agents from overreacting to others through an information bottleneck and aligns the policy with a robust action prior. Empirically, our MIR2 displays even greater resilience against worst-case adversaries than max-min optimization in StarCraft II, Multi-agent Mujoco and rendezvous. Our superiority is consistent when deployed in challenging real-world robot swarm control scenario. See code and demo videos in Supplementary Materials.

The growing prominence of machine learning (ML) and its widespread adoption across industries underscore the need for replicable research [Wagstaff, 2012, Pineau et al., 2021]. Many scientific fields have suffered from this same inability to reproduce the results of published studies [Begley and Ellis, 2012]. Replicability in ML requires not only the ability to reproduce published results [Wagstaff, 2012], as may be partially addressed by sharing code and data [Stodden et al., 2014], but also consistency in the results obtained from successive deployments of an ML algorithm in the same environment. However, the inherent variability and randomness present in ML pose challenges to achieving replicability, as these factors may cause significant variations in results. Building upon foundations of algorithmic stability [Bousquet and Elisseeff, 2002], recent work in learning theory has established rigorous definitions for the study of supervised learning [Impagliazzo et al., 2022] and bandit algorithms [Esfandiari et al., 2023a] that are provably replicable, meaning that algorithms produce identical outputs (with high probability) when executed on distinct data samples from the same underlying distribution. However, these results have not been extended to the study of control problems such as reinforcement learning (RL), that have long been known to suffer from stability issues [White and Eldeib, 1994, Mannor et al., 2004, Islam et al., 2017, Henderson et al., 2018]. These stability issues have already sparked research into robustness for control problems including RL [Khalil et al., 1996, Nilim and Ghaoui, 2005, Iyengar, 2005]. Non-deterministic environments and evaluation benchmarks, the randomness of the exploration process, and the sequential interaction of an RL agent with the environment all complicate the ability to make RL replicable. Our work is orthogonal to that of the robustness literature and our goal is not to reduce the effect of these inherent characteristics, such as by decreasing the amount of exploration that an agent performs, but to develop replicable RL algorithms that support these characteristics.