NeurIPS 2023 Papers

While policy optimization algorithms have played an important role in recent empirical success of Reinforcement Learning (RL), the existing theoretical understanding of policy optimization remains rather limited---they are either restricted to tabular MDPs or suffer from highly suboptimal sample complexity, especial in online RL where exploration is necessary. This paper proposes a simple efficient policy optimization framework---Optimistic NPG for online RL. Optimistic NPG can be viewed as simply combining of the classic natural policy gradient (NPG) algorithm [Kakade, 2001] with optimistic policy evaluation subroutines to encourage exploration. For $d$-dimensional linear MDPs, Optimistic NPG is computationally efficient, and learns an $\epsilon$-optimal policy within $\tilde{\mathcal{O}}(d^2/\epsilon^3)$ samples, which is the first computationally efficient algorithm whose sample complexity has the optimal dimension dependence $\tilde{\Theta}(d^2)$. It also improves over state-of-the-art results of policy optimization algorithms [Zanette et al., 2021] by a factor of $d$. For general function approximation that subsumes linear MDPs, Optimistic NPG, to our best knowledge, is also the first policy optimization algorithm that achieves the polynomial sample complexity for learning near-optimal policies.

One of the grand enduring goals of AI is to create generalist agents that can learn multiple different tasks from diverse data via multitask learning (MTL). However, in practice, applying gradient descent (GD) on the average loss across all tasks may yield poor multitask performance due to severe under-optimization of certain tasks. Previous approaches that manipulate task gradients for a more balanced loss decrease require storing and computing all task gradients ($\mathcal{O}(k)$ space and time where $k$ is the number of tasks), limiting their use in large-scale scenarios. In this work, we introduce Fast Adaptive Multitask Optimization (FAMO), a dynamic weighting method that decreases task losses in a balanced way using $\mathcal{O}(1)$ space and time. We conduct an extensive set of experiments covering multi-task supervised and reinforcement learning problems. Our results indicate that FAMO achieves comparable or superior performance to state-of-the-art gradient manipulation techniques while offering significant improvements in space and computational efficiency. Code is available at \url{https://github.com/Cranial-XIX/FAMO}.

Modern policy optimization methods in reinforcement learning, such as TRPO and PPO, owe their success to the use of parameterized policies. However, while theoretical guarantees have been established for this class of algorithms, especially in the tabular setting, the use of general parameterization schemes remains mostly unjustified. In this work, we introduce a novel framework for policy optimization based on mirror descent that naturally accommodates general parameterizations. The policy class induced by our scheme recovers known classes, e.g., softmax, and generates new ones depending on the choice of mirror map. Using our framework, we obtain the first result that guarantees linear convergence for a policy-gradient-based method involving general parameterization. To demonstrate the ability of our framework to accommodate general parameterization schemes, we provide its sample complexity when using shallow neural networks, show that it represents an improvement upon the previous best results, and empirically validate the effectiveness of our theoretical claims on classic control tasks.

Reinforcement learning presents an attractive paradigm to reason about several distinct aspects of sequential decision making, such as specifying complex goals, planning future observations and actions, and critiquing their utilities. However, the combined integration of these capabilities poses competing algorithmic challenges in retaining maximal expressivity while allowing for flexibility in modeling choices for efficient learning and inference. We present Decision Stacks, a generative framework that decomposes goal-conditioned policy agents into 3 generative modules. These modules simulate the temporal evolution of observations, rewards, and actions via independent generative models that can be learned in parallel via teacher forcing. Our framework guarantees both expressivity and flexibility in designing individual modules to account for key factors such as architectural bias, optimization objective and dynamics, transferrability across domains, and inference speed. Our empirical results demonstrate the effectiveness of Decision Stacks for offline policy optimization for several MDP and POMDP environments, outperforming existing methods and enabling flexible generative decision making.

We introduce a new approach for computing optimal equilibria via learning in games. It applies to extensive-form settings with any number of players, including mechanism design, information design, and solution concepts such as correlated, communication, and certification equilibria. We observe that optimal equilibria are minimax equilibrium strategies of a player in an extensive-form zero-sum game. This reformulation allows to apply techniques for learning in zero-sum games, yielding the first learning dynamics that converge to optimal equilibria, not only in empirical averages, but also in iterates. We demonstrate the practical scalability and flexibility of our approach by attaining state-of-the-art performance in benchmark tabular games, and by computing an optimal mechanism for a sequential auction design problem using deep reinforcement learning.

To make reinforcement learning more sample efficient, we need better credit assignment methods that measure an action’s influence on future rewards. Building upon Hindsight Credit Assignment (HCA), we introduce Counterfactual Contribution Analysis (COCOA), a new family of model-based credit assignment algorithms. Our algorithms achieve precise credit assignment by measuring the contribution of actions upon obtaining subsequent rewards, by quantifying a counterfactual query: ‘Would the agent still have reached this reward if it had taken another action?’. We show that measuring contributions w.r.t. rewarding _states_, as is done in HCA, results in spurious estimates of contributions, causing HCA to degrade towards the high-variance REINFORCE estimator in many relevant environments. Instead, we measure contributions w.r.t. rewards or learned representations of the rewarding objects, resulting in gradient estimates with lower variance. We run experiments on a suite of problems specifically designed to evaluate long-term credit assignment capabilities. By using dynamic programming, we measure ground-truth policy gradients and show that the improved performance of our new model-based credit assignment methods is due to lower bias and variance compared to HCA and common baselines. Our results demonstrate how modeling action contributions towards rewarding outcomes can be leveraged for credit assignment, opening a new path towards sample-efficient reinforcement learning.

Visual navigation has been widely studied under the assumption that there may be several clear routes to reach the goal. However, in more practical scenarios such as a house with several messy rooms, there may not. Interactive Navigation (InterNav) considers agents navigating to their goals more effectively with object interactions, posing new challenges of learning interaction dynamics and extra action space. Previous works learn single vision-to-action policy with the guidance of designed representations. However, the causality between actions and outcomes is prone to be confounded when the attributes of obstacles are diverse and hard to measure. Learning policy for long-term action planning in complex scenes also leads to extensive inefficient exploration. In this paper, we introduce a causal diagram of InterNav clarifying the confounding bias caused by obstacles. To address the problem, we propose a multi-policy model that enables the exploration of counterfactual interactions as well as reduces unnecessary exploration. We develop a large-scale dataset containing 600k task episodes in 12k multi-room scenes based on the ProcTHOR simulator and showcase the effectiveness of our method with the evaluations on our dataset.

The applications of direct policy search in reinforcement learning and continuous control have received increasing attention. In this work, we present novel theoretical results on the complexity of derivative-free policy optimization on an important class of robust control tasks, namely the structured $H_\infty$ synthesis with static output feedback. Optimal $H_\infty$ synthesis under structural constraints leads to a constrained nonconvex nonsmooth problem and is typically addressed using subgradient-based policy search techniques that are built upon the concept of Goldstein subdifferential or other notions of enlarged subdifferential. In this paper, we study the complexity of finding $(\delta,\epsilon)$-stationary points for such nonsmooth robust control design tasks using policy optimization methods which can only access the zeroth-order oracle (i.e. the $H_\infty$ norm of the closed-loop system). First, we study the exact oracle setting and identify the coerciveness of the cost function to prove high-probability feasibility/complexity bounds for derivative-free policy optimization on this problem. Next, we derive a sample complexity result for the multi-input multi-output (MIMO) $H_\infty$-norm estimation. We combine this with our analysis to obtain the first sample complexity of model-free, trajectory-based, zeroth-order policy optimization on finding $(\delta,\epsilon)$-stationary points for structured $H_\infty$ control. Numerical results are also provided to demonstrate our theory.

Existing approaches for improving generalization in deep reinforcement learning (RL) have mostly focused on representation learning, neglecting RL-specific aspects such as exploration. We hypothesize that the agent's exploration strategy plays a key role in its ability to generalize to new environments. Through a series of experiments in a tabular contextual MDP, we show that exploration is helpful not only for efficiently finding the optimal policy for the training environments but also for acquiring knowledge that helps decision making in unseen environments. Based on these observations, we propose EDE: Exploration via Distributional Ensemble, a method that encourages the exploration of states with high epistemic uncertainty through an ensemble of Q-value distributions. The proposed algorithm is the first value-based approach to achieve strong performance on both Procgen and Crafter, two benchmarks for generalization in RL with high-dimensional observations. The open-sourced implementation can be found at https://github.com/facebookresearch/ede.

Deep reinforcement learning (DRL) has led to a wide range of advances in sequential decision-making tasks. However, the complexity of neural network policies makes it difficult to understand and deploy with limited computational resources. Currently, employing compact symbolic expressions as symbolic policies is a promising strategy to obtain simple and interpretable policies. Previous symbolic policy methods usually involve complex training processes and pre-trained neural network policies, which are inefficient and limit the application of symbolic policies. In this paper, we propose an efficient gradient-based learning method named Efficient Symbolic Policy Learning (ESPL) that learns the symbolic policy from scratch in an end-to-end way. We introduce a symbolic network as the search space and employ a path selector to find the compact symbolic policy. By doing so we represent the policy with a differentiable symbolic expression and train it in an off-policy manner which further improves the efficiency. In addition, in contrast with previous symbolic policies which only work in single-task RL because of complexity, we expand ESPL on meta-RL to generate symbolic policies for unseen tasks. Experimentally, we show that our approach generates symbolic policies with higher performance and greatly improves data efficiency for single-task RL. In meta-RL, we demonstrate that compared with neural network policies the proposed symbolic policy achieves higher performance and efficiency and shows the potential to be interpretable.

In reinforcement learning (RL), representation learning is a proven tool for complex image-based tasks, but is often overlooked for environments with low-level states, such as physical control problems. This paper introduces SALE, a novel approach for learning embeddings that model the nuanced interaction between state and action, enabling effective representation learning from low-level states. We extensively study the design space of these embeddings and highlight important design considerations. We integrate SALE and an adaptation of checkpoints for RL into TD3 to form the TD7 algorithm, which significantly outperforms existing continuous control algorithms. On OpenAI gym benchmark tasks, TD7 has an average performance gain of 276.7% and 50.7% over TD3 at 300k and 5M time steps, respectively, and works in both the online and offline settings.

Monte-Carlo path tracing is a powerful technique for realistic image synthesis but suffers from high levels of noise at low sample counts, limiting its use in real-time applications. To address this, we propose a framework with end-to-end training of a sampling importance network, a latent space encoder network, and a denoiser network. Our approach uses reinforcement learning to optimize the sampling importance network, thus avoiding explicit numerically approximated gradients. Our method does not aggregate the sampled values per pixel by averaging but keeps all sampled values which are then fed into the latent space encoder. The encoder replaces handcrafted spatiotemporal heuristics by learned representations in a latent space. Finally, a neural denoiser is trained to refine the output image. Our approach increases visual quality on several challenging datasets and reduces rendering times for equal quality by a factor of 1.6x compared to the previous state-of-the-art, making it a promising solution for real-time applications.

Skill discovery has shown significant progress in unsupervised reinforcement learning. This approach enables the discovery of a wide range of skills without any extrinsic reward, which can be effectively combined to tackle complex tasks. However, such unsupervised skill learning has not been well applied to multi-agent reinforcement learning (MARL) due to two primary challenges. One is how to learn skills not only for the individual agents but also for the entire team, and the other is how to coordinate the skills of different agents to accomplish multi-agent tasks. To address these challenges, we present Hierarchical Multi-Agent Skill Discovery (HMASD), a two-level hierarchical algorithm for discovering both team and individual skills in MARL. The high-level policy employs a transformer structure to realize sequential skill assignment, while the low-level policy learns to discover valuable team and individual skills. We evaluate HMASD on sparse reward multi-agent benchmarks, and the results show that HMASD achieves significant performance improvements compared to strong MARL baselines.

As more non-AI experts use complex AI systems for daily tasks, there has been an increasing effort to develop methods that produce explanations of AI decision making that are understandable by non-AI experts. Towards this effort, leveraging higher-level concepts and producing concept-based explanations have become a popular method. Most concept-based explanations have been developed for classification techniques, and we posit that the few existing methods for sequential decision making are limited in scope. In this work, we first contribute a desiderata for defining ``concepts'' in sequential decision making settings. Additionally, inspired by the Protege Effect which states explaining knowledge often reinforces one's self-learning, we explore how concept-based explanations of an RL agent's decision making can in turn improve the agent's learning rate, as well as improve end-user understanding of the agent's decision making. To this end, we contribute a unified framework, State2Explanation (S2E), that involves learning a joint embedding model between state-action pairs and concept-based explanations, and leveraging such learned model to both (1) inform reward shaping during an agent's training, and (2) provide explanations to end-users at deployment for improved task performance. Our experimental validations, in Connect 4 and Lunar Lander, demonstrate the success of S2E in providing a dual-benefit, successfully informing reward shaping and improving agent learning rate, as well as significantly improving end user task performance at deployment time.

An intelligent conversational agent (a.k.a., chat-bot) could embrace conversational technologies to obtain user preferences online, to overcome inherent limitations of recommender systems trained over the offline historical user behaviors. In this paper, we propose CORE, a new offline-training and online-checking framework to plug a COnversational agent into REcommender systems. Unlike most prior conversational recommendation approaches that systemically combine conversational and recommender parts through a reinforcement learning framework, CORE bridges the conversational agent and recommender system through a unified uncertainty minimization framework, which can be easily applied to any existing recommendation approach. Concretely, CORE treats a recommender system as an offline estimator to produce an estimated relevance score for each item, while CORE regards a conversational agent as an online checker that checks these estimated scores in each online session. We define uncertainty as the sum of unchecked relevance scores. In this regard, the conversational agent acts to minimize uncertainty via querying either attributes or items. Towards uncertainty minimization, we derive the certainty gain of querying each attribute and item, and develop a novel online decision tree algorithm to decide what to query at each turn. Our theoretical analysis reveals the bound of the expected number of turns of CORE in a cold-start setting. Experimental results demonstrate that CORE can be seamlessly employed on a variety of recommendation approaches, and can consistently bring significant improvements in both hot-start and cold-start settings.

We study robust reinforcement learning (RL) with the goal of determining a well-performing policy that is robust against model mismatch between the training simulator and the testing environment. Previous policy-based robust RL algorithms mainly focus on the tabular setting under uncertainty sets that facilitate robust policy evaluation, but are no longer tractable when the number of states scales up. To this end, we propose two novel uncertainty set formulations, one based on double sampling and the other on an integral probability metric. Both make large-scale robust RL tractable even when one only has access to a simulator. We propose a robust natural actor-critic (RNAC) approach that incorporates the new uncertainty sets and employs function approximation. We provide finite-time convergence guarantees for the proposed RNAC algorithm to the optimal robust policy within the function approximation error. Finally, we demonstrate the robust performance of the policy learned by our proposed RNAC approach in multiple MuJoCo environments and a real-world TurtleBot navigation task.

As the capabilities of artificial agents improve, they are being increasingly deployed to service multiple diverse objectives and stakeholders. However, the composition of these objectives is often performed ad hoc, with no clear justification. This paper takes a normative approach to multi-objective agency: from a set of intuitively appealing axioms, it is shown that Markovian aggregation of Markovian reward functions is not possible when the time preference (discount factor) for each objective may vary. It follows that optimal multi-objective agents must admit rewards that are non-Markovian with respect to the individual objectives. To this end, a practical non-Markovian aggregation scheme is proposed, which overcomes the impossibility with only one additional parameter for each objective. This work offers new insights into sequential, multi-objective agency and intertemporal choice, and has practical implications for the design of AI systems deployed to serve multiple generations of principals with varying time preference.

In value-based deep reinforcement learning with replay memories, the batch size parameter specifies how many transitions to sample for each gradient update. Although critical to the learning process, this value is typically not adjusted when proposing new algorithms. In this work we present a broad empirical study that suggests reducing the batch size can result in a number of significant performance gains; this is surprising, as the general tendency when training neural networks is towards larger batch sizes for improved performance. We complement our experimental findings with a set of empirical analyses towards better understanding this phenomenon.

A major challenge in reinforcement learning is to determine which state-action pairs are responsible for future rewards that are delayed. Reward redistribution serves as a solution to re-assign credits for each time step from observed sequences. While the majority of current approaches construct the reward redistribution in an uninterpretable manner, we propose to explicitly model the contributions of state and action from a causal perspective, resulting in an interpretable reward redistribution and preserving policy invariance. In this paper, we start by studying the role of causal generative models in reward redistribution by characterizing the generation of Markovian rewards and trajectory-wise long-term return and further propose a framework, called Generative Return Decomposition (GRD), for policy optimization in delayed reward scenarios. Specifically, GRD first identifies the unobservable Markovian rewards and causal relations in the generative process. Then, GRD makes use of the identified causal generative model to form a compact representation to train policy over the most favorable subspace of the state space of the agent. Theoretically, we show that the unobservable Markovian reward function is identifiable, as well as the underlying causal structure and causal models. Experimental results show that our method outperforms state-of-the-art methods and the provided visualization further demonstrates the interpretability of our method. The project page is located at [https://reedzyd.github.io/GenerativeReturnDecomposition/](https://reedzyd.github.io/GenerativeReturnDecomposition/).