Federated learning (FL) is becoming a major driving force behind machine learning as a service, where customers (clients) collaboratively benefit from shared local updates under the orchestration of the service provider (server). Representing clients' current demands and the server's future demand, local model personalization and global model generalization are separately investigated, as the ill-effects of data heterogeneity enforce the community to focus on one over the other. However, these two seemingly competing goals are of equal importance rather than black and white issues, and should be achieved simultaneously. In this paper, we propose the first algorithm to balance personalization and generalization on top of game theory, dubbed PAGE, which reshapes FL as a co-opetition game between clients and the server. To explore the equilibrium, PAGE further formulates the game as Markov decision processes, and leverages the reinforcement learning algorithm, which simplifies the solving complexity. Extensive experiments on four widespread datasets show that PAGE outperforms state-of-the-art FL baselines in terms of global and local prediction accuracy simultaneously, and the accuracy can be improved by up to 35.20% and 39.91%, respectively. In addition, biased variants of PAGE imply promising adaptiveness to demand shifts in practice.

Unsupervised pre-training strategies have proven to be highly effective in natural language processing and computer vision. Likewise, unsupervised reinforcement learning (RL) holds the promise of discovering a variety of potentially useful behaviors that can accelerate the learning of a wide array of downstream tasks. Previous unsupervised RL approaches have mainly focused on pure exploration and mutual information skill learning. However, despite the previous attempts, making unsupervised RL truly scalable still remains a major open challenge: pure exploration approaches might struggle in complex environments with large state spaces, where covering every possible transition is infeasible, and mutual information skill learning approaches might completely fail to explore the environment due to the lack of incentives. To make unsupervised RL scalable to complex, high-dimensional environments, we propose a novel unsupervised RL objective, which we call Metric-Aware Abstraction (METRA). Our main idea is, instead of directly covering the entire state space, to only cover a compact latent space $Z$ that is metrically connected to the state space $S$ by temporal distances. By learning to move in every direction in the latent space, METRA obtains a tractable set of diverse behaviors that approximately cover the state space, being scalable to high-dimensional environments. Through our experiments in five locomotion and manipulation environments, we demonstrate that METRA can discover a variety of useful behaviors even in complex, pixel-based environments, being the first unsupervised RL method that discovers diverse locomotion behaviors in pixel-based Quadruped and Humanoid. Our code and videos are available at https://seohong.me/projects/metra/

Reliability quantification of deep reinforcement learning (DRL)-based control is a significant challenge for the practical application of artificial intelligence (AI) in safety-critical systems. This study proposes a method for quantifying the reliability of DRL-based control. First, an existing method, random noise distillation, was applied to the reliability evaluation to clarify the issues to be solved. Second, a novel method for reliability quantification was proposed to solve these issues. The reliability is quantified using two neural networks: reference and evaluator. They have the same structure with the same initial parameters. The outputs of the two networks were the same before training. During training, the evaluator network parameters were updated to maximize the difference between the reference and evaluator networks for trained data. Thus, the reliability of the DRL-based control for a state can be evaluated based on the difference in output between the two networks. The proposed method was applied to DQN-based control as an example of a simple task, and its effectiveness was demonstrated. Finally, the proposed method was applied to the problem of switching trained models depending on the state. Con-sequently, the performance of the DRL-based control was improved by switching the trained models according to their reliability.

Coverage path planning (CPP) is the problem of finding a path that covers the entire free space of a confined area, with applications ranging from robotic lawn mowing and vacuum cleaning, to demining and search-and-rescue tasks. While offline methods can find provably complete, and in some cases optimal, paths for known environments, their value is limited in online scenarios where the environment is not known beforehand. In this case, the path needs to be planned online while mapping the environment. We investigate how suitable reinforcement learning is for this challenging problem, and analyze the involved components required to efficiently learn coverage paths, such as action space, input feature representation, neural network architecture, and reward function. Compared to existing classical methods, this approach allows for a flexible path space, and enables the agent to adapt to specific environment dynamics. In addition to local sensory inputs for acting on short-term obstacle detections, we propose to use egocentric maps in multiple scales based on frontiers. This allows the agent to plan a long-term path in large-scale environments with feasible computational and memory complexity. Furthermore, we propose a novel total variation reward term for guiding the agent not to leave small holes of non-covered free space. To validate the effectiveness of our approach, we perform extensive experiments in simulation with a 2D ranging sensor on different variations of the CPP problem, surpassing the performance of both previous RL-based approaches and highly specialized methods.

Continuum robots have gained widespread popularity due to their inherent compliance and flexibility, particularly their adjustable levels of stiffness for various application scenarios. Despite efforts to dynamic modeling and control synthesis over the past decade, few studies have incorporated stiffness regulation into their feedback control design; however, this is one of the initial motivations to develop continuum robots. This paper addresses the crucial challenge of controlling both the position and stiffness of underactuated continuum robots actuated by antagonistic tendons. We begin by presenting a rigid-link dynamical model that can analyze the open-loop stiffening of tendon-driven continuum robots. Based on this model, we propose a novel passivity-based position-and-stiffness controller that adheres to the non-negative tension constraint. Comprehensive experiments on our continuum robot validate the theoretical results and demonstrate the efficacy and precision of this approach.

Generalizing policies across different domains with dynamics mismatch poses a significant challenge in reinforcement learning. For example, a robot learns the policy in a simulator, but when it is deployed in the real world, the dynamics of the environment may be different. Given the source and target domain with dynamics mismatch, we consider the online dynamics adaptation problem, in which case the agent can access sufficient source domain data while online interactions with the target domain are limited. Existing research has attempted to solve the problem from the dynamics discrepancy perspective. In this work, we reveal the limitations of these methods and explore the problem from the value difference perspective via a novel insight on the value consistency across domains. Specifically, we present the Value-Guided Data Filtering (VGDF) algorithm, which selectively shares transitions from the source domain based on the proximity of paired value targets across the two domains. Empirical results on various environments with kinematic and morphology shifts demonstrate that our method achieves superior performance compared to prior approaches.

Adversarial examples can be useful for identifying vulnerabilities in AI systems before they are deployed. In reinforcement learning (RL), adversarial policies can be developed by training an adversarial agent to minimize a target agent's rewards. Prior work has studied black-box versions of these attacks where the adversary only observes the world state and treats the target agent as any other part of the environment. However, this does not take into account additional structure in the problem. In this work, we study white-box adversarial policies and show that having access to a target agent's internal state can be useful for identifying its vulnerabilities. We make two contributions. (1) We introduce white-box adversarial policies where an attacker observes both a target's internal state and the world state at each timestep. We formulate ways of using these policies to attack agents in 2-player games and text-generating language models. (2) We demonstrate that these policies can achieve higher initial and asymptotic performance against a target agent than black-box controls. Code is available at https://github.com/thestephencasper/lm_white_box_attacks

The online advertising market, with its thousands of auctions run per second, presents a daunting challenge for advertisers who wish to optimize their spend under a budget constraint. Thus, advertising platforms typically provide automated agents to their customers, which act on their behalf to bid for impression opportunities in real time at scale. Because these proxy agents are owned by the platform but use advertiser funds to operate, there is a strong practical need to balance reliability and explainability of the agent with optimizing power. We propose a generalizable approach to optimizing bidding policies in production environments by learning from real data using offline reinforcement learning. This approach can be used to optimize any differentiable base policy (practically, a heuristic policy based on principles which the advertiser can easily understand), and only requires data generated by the base policy itself. We use a hybrid agent architecture that combines arbitrary base policies with deep neural networks, where only the optimized base policy parameters are eventually deployed, and the neural network part is discarded after training. We demonstrate that such an architecture achieves statistically significant performance gains in both simulated and at-scale production bidding environments. Our approach does not incur additional infrastructure, safety, or explainability costs, as it directly optimizes parameters of existing production routines without replacing them with black box-style models like neural networks.

Personalization of playlists is a common feature in music streaming services, but conventional techniques, such as collaborative filtering, rely on explicit assumptions regarding content quality to learn how to make recommendations. Such assumptions often result in misalignment between offline model objectives and online user satisfaction metrics. In this paper, we present a reinforcement learning framework that solves for such limitations by directly optimizing for user satisfaction metrics via the use of a simulated playlist-generation environment. Using this simulator we develop and train a modified Deep Q-Network, the action head DQN (AH-DQN), in a manner that addresses the challenges imposed by the large state and action space of our RL formulation. The resulting policy is capable of making recommendations from large and dynamic sets of candidate items with the expectation of maximizing consumption metrics. We analyze and evaluate agents offline via simulations that use environment models trained on both public and proprietary streaming datasets. We show how these agents lead to better user-satisfaction metrics compared to baseline methods during online A/B tests. Finally, we demonstrate that performance assessments produced from our simulator are strongly correlated with observed online metric results.

In this paper, we study the fundamental statistical efficiency of Reinforcement Learning in Mean-Field Control (MFC) and Mean-Field Game (MFG) with general model-based function approximation. We introduce a new concept called Mean-Field Model-Based Eluder Dimension (MF-MBED), which characterizes the inherent complexity of mean-field model classes. We show that low MF-MBED subsumes a rich family of Mean-Field RL problems. Additionally, we propose algorithms based on maximal likelihood estimation, which can return an $\epsilon$-optimal policy for MFC or an $\epsilon$-Nash Equilibrium policy for MFG, with sample complexity polynomial w.r.t. relevant parameters and independent of the number of states, actions and agents. Compared with previous works, our results only require the minimal assumptions including realizability and Lipschitz continuity.