We study the exploration problem with approximate linear action-value functions in episodic reinforcement learning under the notion of low inherent Bellman error, a condition normally employed to show convergence of approximate value iteration. First we relate this condition to other common frameworks and show that it is strictly more general than the low rank (or linear) MDP assumption of prior work. Second we provide an algorithm with a high probability regret bound $\widetilde O(\sum_{t=1}^H d_t \sqrt{K} + \sum_{t=1}^H \sqrt{d_t} \IBE K)$ where $H$ is the horizon, $K$ is the number of episodes, $\IBE$ is the value if the inherent Bellman error and $d_t$ is the feature dimension at timestep $t$. In addition, we show that the result is unimprovable beyond constants and logs by showing a matching lower bound. This has two important consequences: 1) it shows that exploration is possible using only \emph{batch assumptions} with an algorithm that achieves the optimal statistical rate for the setting we consider, which is more general than prior work on low-rank MDPs 2) the lack of closedness (measured by the inherent Bellman error) is only amplified by $\sqrt{d_t}$ despite working in the online setting. Finally, the algorithm reduces to the celebrated \textsc{LinUCB} when $H=1$ but with a different choice of the exploration parameter that allows handling misspecified contextual linear bandits. While computational tractability questions remain open for the MDP setting, this enriches the class of MDPs with a linear representation for the action-value function where statistically efficient reinforcement learning is possible.
School of Operations Research and Information Engineering, Cornell University（康奈尔大学运营研究与信息工程学院）
We develop a framework for designing simple and efficient policies for a family of online allocation and pricing problems, that includes online packing, budget-constrained probing, dynamic pricing, and online contextual bandits with knapsacks. In each case, we evaluate the performance of our policies in terms of their regret (i.e., additive gap) relative to an offline controller that is endowed with more information than the online controller. Our framework is based on Bellman Inequalities, which decompose the loss of an algorithm into two distinct sources of error: (1) arising from computatio…
The impact of softmax on the value function itself in reinforcement learning (RL) is often viewed as problematic because it leads to sub-optimal value (or Q) functions and interferes with the contraction properties of the Bellman operator. Surprisingly, despite these concerns, and independent of its effect on exploration, the softmax Bellman operator when combined with Deep Q-learning, leads to Q-functions with superior policies in practice, even outperforming its double Q-learning counterpart. To better understand how and why this occurs, we revisit theoretical properties of the softmax Bellm…
Recently, \citet{SuttonMW15} introduced the emphatic temporal differences (ETD) algorithm for off-policy evaluation in Markov decision processes. In this short note, we show that the projected fixed-point equation that underlies ETD involves a contraction operator, with a $\sqrtγ$-contraction modulus (where $γ$ is the discount factor). This allows us to provide error bounds on the approximation error of ETD. To our knowledge, these are the first error bounds for an off-policy evaluation algorithm under general target and behavior policies.
Finding the minimal structural assumptions that empower sample-efficient learning is one of the most important research directions in Reinforcement Learning (RL). This paper advances our understanding of this fundamental question by introducing a new complexity measure -- Bellman Eluder (BE) dimension. We show that the family of RL problems of low BE dimension is remarkably rich, which subsumes a vast majority of existing tractable RL problems including but not limited to tabular MDPs, linear MDPs, reactive POMDPs, low Bellman rank problems as well as low Eluder dimension problems. This paper …
UC Berkeley（加州大学伯克利分校）；Massachusetts Institute of Technology（麻省理工学院）
We propose and analyze a reinforcement learning principle that approximates the Bellman equations by enforcing their validity only along an user-defined space of test functions. Focusing on applications to model-free offline RL with function approximation, we exploit this principle to derive confidence intervals for off-policy evaluation, as well as to optimize over policies within a prescribed policy class. We prove an oracle inequality on our policy optimization procedure in terms of a trade-off between the value and uncertainty of an arbitrary comparator policy. Different choices of test fu…
Sequential decision making in the presence of uncertainty and stochastic dynamics gives rise to distributions over state/action trajectories in reinforcement learning (RL) and optimal control problems. This observation has led to a variety of connections between RL and inference in probabilistic graphical models (PGMs). Here we explore a different dimension to this relationship, examining reinforcement learning using the tools and abstractions of statistical physics. The central object in the statistical physics abstraction is the idea of a partition function $\mathcal{Z}$, and here we constru…
We investigate projection methods, for evaluating a linear approximation of the value function of a policy in a Markov Decision Process context. We consider two popular approaches, the one-step Temporal Difference fix-point computation (TD(0)) and the Bellman Residual (BR) minimization. We describe examples, where each method outperforms the other. We highlight a simple relation between the objective function they minimize, and show that while BR enjoys a performance guarantee, TD(0) does not in general. We then propose a unified view in terms of oblique projections of the Bellman equation, wh…
Université de Lorraine（洛林大学）；CNRS（法国国家科学研究中心）；Univ. Lille（里尔大学）；Centrale Lille（里尔中央理工学院）；Inria（法国国家信息与自动化研究所）；Google DeepMind（谷歌DeepMind）
This paper aims at theoretically and empirically comparing two standard optimization criteria for Reinforcement Learning: i) maximization of the mean value and ii) minimization of the Bellman residual. For that purpose, we place ourselves in the framework of policy search algorithms, that are usually designed to maximize the mean value, and derive a method that minimizes the residual $\|T_* v_π- v_π\|_{1,ν}$ over policies. A theoretical analysis shows how good this proxy is to policy optimization, and notably that it is better than its value-based counterpart. We also propose experiments on ra…
One of the main challenges in model-based reinforcement learning (RL) is to decide which aspects of the environment should be modeled. The value-equivalence (VE) principle proposes a simple answer to this question: a model should capture the aspects of the environment that are relevant for value-based planning. Technically, VE distinguishes models based on a set of policies and a set of functions: a model is said to be VE to the environment if the Bellman operators it induces for the policies yield the correct result when applied to the functions. As the number of policies and functions increa…
State-action value functions (i.e., Q-values) are ubiquitous in reinforcement learning (RL), giving rise to popular algorithms such as SARSA and Q-learning. We propose a new notion of action value defined by a Gaussian smoothed version of the expected Q-value. We show that such smoothed Q-values still satisfy a Bellman equation, making them learnable from experience sampled from an environment. Moreover, the gradients of expected reward with respect to the mean and covariance of a parameterized Gaussian policy can be recovered from the gradient and Hessian of the smoothed Q-value function. Bas…
We introduce a new algorithm for multi-objective reinforcement learning (MORL) with linear preferences, with the goal of enabling few-shot adaptation to new tasks. In MORL, the aim is to learn policies over multiple competing objectives whose relative importance (preferences) is unknown to the agent. While this alleviates dependence on scalar reward design, the expected return of a policy can change significantly with varying preferences, making it challenging to learn a single model to produce optimal policies under different preference conditions. We propose a generalized version of the Bell…
While maximizing expected return is the goal in most reinforcement learning approaches, risk-sensitive objectives such as conditional value at risk (CVaR) are more suitable for many high-stakes applications. However, relatively little is known about how to explore to quickly learn policies with good CVaR. In this paper, we present the first algorithm for sample-efficient learning of CVaR-optimal policies in Markov decision processes based on the optimism in the face of uncertainty principle. This method relies on a novel optimistic version of the distributional Bellman operator that moves prob…
University of Illinois at Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）
We prove performance guarantees of two algorithms for approximating $Q^\star$ in batch reinforcement learning. Compared to classical iterative methods such as Fitted Q-Iteration---whose performance loss incurs quadratic dependence on horizon---these methods estimate (some forms of) the Bellman error and enjoy linear-in-horizon error propagation, a property established for the first time for algorithms that rely solely on batch data and output stationary policies. One of the algorithms uses a novel and explicit importance-weighting correction to overcome the infamous "double sampling" difficult…
We present a unified framework for learning continuous control policies using backpropagation. It supports stochastic control by treating stochasticity in the Bellman equation as a deterministic function of exogenous noise. The product is a spectrum of general policy gradient algorithms that range from model-free methods with value functions to model-based methods without value functions. We use learned models but only require observations from the environment in- stead of observations from model-predicted trajectories, minimizing the impact of compounded model errors. We apply these algorithm…
McGill University（麦吉尔大学）；Mila – Quebec AI Institute（米拉-魁北克人工智能研究所）；Google Brain（谷歌大脑）
Continuous-time reinforcement learning offers an appealing formalism for describing control problems in which the passage of time is not naturally divided into discrete increments. Here we consider the problem of predicting the distribution of returns obtained by an agent interacting in a continuous-time, stochastic environment. Accurate return predictions have proven useful for determining optimal policies for risk-sensitive control, learning state representations, multiagent coordination, and more. We begin by establishing the distributional analogue of the Hamilton-Jacobi-Bellman (HJB) equa…
We consider infinite-horizon $γ$-discounted Markov Decision Processes, for which it is known that there exists a stationary optimal policy. We consider the algorithm Value Iteration and the sequence of policies $π_1,...,π_k$ it implicitely generates until some iteration $k$. We provide performance bounds for non-stationary policies involving the last $m$ generated policies that reduce the state-of-the-art bound for the last stationary policy $π_k$ by a factor $\frac{1-γ}{1-γ^m}$. In particular, the use of non-stationary policies allows to reduce the usual asymptotic performance bounds of Value…
In many real-life reinforcement learning (RL) problems, deploying new policies is costly. In those scenarios, algorithms must solve exploration (which requires adaptivity) while switching the deployed policy sparsely (which limits adaptivity). In this paper, we go beyond the existing state-of-the-art on this problem that focused on linear Markov Decision Processes (MDPs) by considering linear Bellman-complete MDPs with low inherent Bellman error. We propose the ELEANOR-LowSwitching algorithm that achieves the near-optimal regret with a switching cost logarithmic in the number of episodes and l…
We consider a new form of reinforcement learning (RL) that is based on opportunities to directly learn the optimal control policy and a general Markov decision process (MDP) framework devised to support these opportunities. Derivations of general classes of our control-based RL methods are presented, together with forms of exploration and exploitation in learning and applying the optimal control policy over time. Our general MDP framework extends the classical Bellman operator and optimality criteria by generalizing the definition and scope of a policy for any given state. We establish the con…
Vector-Valued Distributional Reinforcement Learning Policy Evaluation: A Hilbert Space Embedding Approach
Department of Statistics, University of Illinois Urbana-Champaign（统计系，伊利诺伊大学厄巴纳-香槟分校）；Department of Bioinformatics and Biostatistics, University of Louisville（生物信息学与生物统计学系，路易斯维尔大学）
This paper introduces new optimality-preserving operators on Q-functions. We first describe an operator for tabular representations, the consistent Bellman operator, which incorporates a notion of local policy consistency. We show that this local consistency leads to an increase in the action gap at each state; increasing this gap, we argue, mitigates the undesirable effects of approximation and estimation errors on the induced greedy policies. This operator can also be applied to discretized continuous space and time problems, and we provide empirical results evidencing superior performance i…
Technion（以色列理工学院）；National University of Singapore（新加坡国立大学）
We present an efficient robust value iteration for \texttt{s}-rectangular robust Markov Decision Processes (MDPs) with a time complexity comparable to standard (non-robust) MDPs which is significantly faster than any existing method. We do so by deriving the optimal robust Bellman operator in concrete forms using our $L_p$ water filling lemma. We unveil the exact form of the optimal policies, which turn out to be novel threshold policies with the probability of playing an action proportional to its advantage.
Finite-Time Convergence of Single-Trajectory Chi-Square Robust Q-Learning With Linear Function Approximation
ECE and CSL University of Illinois Urbana-Champaign（电子与计算机工程系和计算机科学实验室，伊利诺伊大学厄巴纳-香槟分校）；Computing and Mathematical Sciences California Institute of Technology（计算与数学科学，加州理工学院）；ECE, CSL, and NCSA University of Illinois Urbana-Champaign（电子与计算机工程系、计算机科学实验室和网络与计算科学中心，伊利诺伊大学厄巴纳-香槟分校）
MIIT Key Laboratory of Pattern Analysis and Machine Intelligence（工业和信息化部模式分析与机器智能重点实验室）
Most of the policy evaluation algorithms are based on the theories of Bellman Expectation and Optimality Equation, which derive two popular approaches - Policy Iteration (PI) and Value Iteration (VI). However, multi-step bootstrapping is often at cross-purposes with and off-policy learning in PI-based methods due to the large variance of multi-step off-policy correction. In contrast, VI-based methods are naturally off-policy but subject to one-step learning.In this paper, we deduce a novel multi-step Bellman Optimality Equation by utilizing a latent structure of multi-step bootstrapping with t…
Reinforcement learning in a multi agent system is difficult because these systems are inherently non-stationary in nature. In such a case, identifying the type of the opposite agent is crucial and can help us address this non-stationary environment. We have investigated if we can employ some probing policies which help us better identify the type of the other agent in the environment. We've made a simplifying assumption that the other agent has a stationary policy that our probing policy is trying to approximate. Our work extends Environmental Probing Interaction Policy framework to handle mul…
We consider the stochastic shortest path planning problem in MDPs, i.e., the problem of designing policies that ensure reaching a goal state from a given initial state with minimum accrued cost. In order to account for rare but important realizations of the system, we consider a nested dynamic coherent risk total cost functional rather than the conventional risk-neutral total expected cost. Under some assumptions, we show that optimal, stationary, Markovian policies exist and can be found via a special Bellman's equation. We propose a computational technique based on difference convex programs…
We provide a framework for incorporating robustness -- to perturbations in the transition dynamics which we refer to as model misspecification -- into continuous control Reinforcement Learning (RL) algorithms. We specifically focus on incorporating robustness into a state-of-the-art continuous control RL algorithm called Maximum a-posteriori Policy Optimization (MPO). We achieve this by learning a policy that optimizes for a worst case expected return objective and derive a corresponding robust entropy-regularized Bellman contraction operator. In addition, we introduce a less conservative, sof…
Reinforcement Learning with Imperfect Transition Predictions: A Bellman-Jensen Approach
Institute for Interdisciplinary Information Sciences, Tsinghua University（清华大学交叉信息研究院）；Edwardson School of Industrial Engineering, Purdue University（普渡大学爱德华森工业工程学院）；School of Data Science, The Chinese University of Hong Kong (Shenzhen)（香港中文大学（深圳）数据科学学院）；School of Science and Engineering, The Chinese University of Hong Kong (Shenzhen)（香港中文大学（深圳）理工学院）；Computing & Mathematical Sciences, Caltech（加州理工学院计算与数学科学系）
Google Research（谷歌研究院）；Tel Aviv University（特拉维夫大学）；Courant Institute of Mathematical Sciences（柯朗数学科学研究所）；Cornell University（康奈尔大学）
Myopic exploration policies such as epsilon-greedy, softmax, or Gaussian noise fail to explore efficiently in some reinforcement learning tasks and yet, they perform well in many others. In fact, in practice, they are often selected as the top choices, due to their simplicity. But, for what tasks do such policies succeed? Can we give theoretical guarantees for their favorable performance? These crucial questions have been scarcely investigated, despite the prominent practical importance of these policies. This paper presents a theoretical analysis of such policies and provides the first regret…
