NeurIPS 2024 论文列表

Inverse Reinforcement Learning (IRL) deals with the problem of deducing a reward function that explains the behavior of an expert agent who is assumed to act *optimally* in an underlying unknown task. Recent works have studied the IRL problem from the perspective of recovering the *feasible reward set*, i.e., the class of reward functions that are compatible with a unique optimal expert. However, in several problems of interest it is possible to observe the behavior of multiple experts with different degree of optimality (e.g., racing drivers whose skills ranges from amateurs to professionals). For this reason, in this work, we focus on the reconstruction of the feasible reward set when, in addition to demonstrations from the optimal expert, we observe the behavior of multiple *sub-optimal experts*. Given this problem, we first study the theoretical properties showing that the presence of multiple sub-optimal experts, in addition to the optimal one, can significantly shrink the set of compatible rewards, ultimately mitigating the inherent ambiguity of IRL. Furthermore, we study the statistical complexity of estimating the feasible reward set with a generative model and analyze a uniform sampling algorithm that turns out to be minimax optimal whenever the sub-optimal experts' performance level is sufficiently close to that of the optimal expert.

We study reinforcement learning with _multinomial logistic_ (MNL) function approximation where the underlying transition probability kernel of the _Markov decision processes_ (MDPs) is parametrized by an unknown transition core with features of state and action. For the finite horizon episodic setting with inhomogeneous state transitions, we propose provably efficient algorithms with randomized exploration having frequentist regret guarantees. For our first algorithm, $\texttt{RRL-MNL}$, we adapt optimistic sampling to ensure the optimism of the estimated value function with sufficient frequency and establish that $\texttt{RRL-MNL}$ is both _statistically_ and _computationally_ efficient, achieving a $\tilde{\mathcal{O}}(\kappa^{-1} d^{\frac{3}{2}} H^{\frac{3}{2}} \sqrt{T})$ frequentist regret bound with constant-time computational cost per episode. Here, $d$ is the dimension of the transition core, $H$ is the horizon length, $T$ is the total number of steps, and $\kappa$ is a problem-dependent constant. Despite the simplicity and practicality of $\texttt{RRL-MNL}$, its regret bound scales with $\kappa^{-1}$, which is potentially large in the worst case. To improve the dependence on $\kappa^{-1}$, we propose $\texttt{ORRL-MNL}$, which estimates the value function using local gradient information of the MNL transition model. We show that its frequentist regret bound is $\tilde{\mathcal{O}}(d^{\frac{3}{2}} H^{\frac{3}{2}} \sqrt{T} + \kappa^{-1} d^2 H^2)$. To the best of our knowledge, these are the first randomized RL algorithms for the MNL transition model that achieve both computational and statistical efficiency. Numerical experiments demonstrate the superior performance of the proposed algorithms.

As a prominent category of imitation learning methods, adversarial imitation learning (AIL) has garnered significant practical success powered by neural network approximation. However, existing theoretical studies on AIL are primarily limited to simplified scenarios such as tabular and linear function approximation and involve complex algorithmic designs that hinder practical implementation, highlighting a gap between theory and practice. In this paper, we explore the theoretical underpinnings of online AIL with general function approximation. We introduce a new method called optimization-based AIL (OPT-AIL), which centers on performing online optimization for reward functions and optimism-regularized Bellman error minimization for Q-value functions. Theoretically, we prove that OPT-AIL achieves polynomial expert sample complexity and interaction complexity for learning near-expert policies. To our best knowledge, OPT-AIL is the first provably efficient AIL method with general function approximation. Practically, OPT-AIL only requires the approximate optimization of two objectives, thereby facilitating practical implementation. Empirical studies demonstrate that OPT-AIL outperforms previous state-of-the-art deep AIL methods in several challenging tasks.

Offline Reinforcement Learning (RL) suffers from the extrapolation error and value overestimation. From a generalization perspective, this issue can be attributed to the over-generalization of value functions or policies towards out-of-distribution (OOD) actions. Significant efforts have been devoted to mitigating such generalization, and recent in-sample learning approaches have further succeeded in entirely eschewing it. Nevertheless, we show that mild generalization beyond the dataset can be trusted and leveraged to improve performance under certain conditions. To appropriately exploit generalization in offline RL, we propose Doubly Mild Generalization (DMG), comprising (i) mild action generalization and (ii) mild generalization propagation. The former refers to selecting actions in a close neighborhood of the dataset to maximize the Q values. Even so, the potential erroneous generalization can still be propagated, accumulated, and exacerbated by bootstrapping. In light of this, the latter concept is introduced to mitigate the generalization propagation without impeding the propagation of RL learning signals. Theoretically, DMG guarantees better performance than the in-sample optimal policy in the oracle generalization scenario. Even under worst-case generalization, DMG can still control value overestimation at a certain level and lower bound the performance. Empirically, DMG achieves state-of-the-art performance across Gym-MuJoCo locomotion tasks and challenging AntMaze tasks. Moreover, benefiting from its flexibility in both generalization aspects, DMG enjoys a seamless transition from offline to online learning and attains strong online fine-tuning performance.

Offline reinforcement learning (RL) leverages pre-collected datasets to train optimal policies. Diffusion Q-Learning (DQL), introducing diffusion models as a powerful and expressive policy class, significantly boosts the performance of offline RL. However, its reliance on iterative denoising sampling to generate actions slows down both training and inference. While several recent attempts have tried to accelerate diffusion-QL, the improvement in training and/or inference speed often results in degraded performance. In this paper, we introduce a dual policy approach, Diffusion Trusted Q-Learning (DTQL), which comprises a diffusion policy for pure behavior cloning and a practical one-step policy. We bridge the two polices by a newly introduced diffusion trust region loss. The diffusion policy maintains expressiveness, while the trust region loss directs the one-step policy to explore freely and seek modes within the region defined by the diffusion policy. DTQL eliminates the need for iterative denoising sampling during both training and inference, making it remarkably computationally efficient. We evaluate its effectiveness and algorithmic characteristics against popular Kullback-Leibler (KL) based distillation methods in 2D bandit scenarios and gym tasks. We then show that DTQL could not only outperform other methods on the majority of the D4RL benchmark tasks but also demonstrate efficiency in training and inference speeds. The PyTorch implementation is available at https://github.com/TianyuCodings/Diffusion_Trusted_Q_Learning.

We study offline multitask representation learning in reinforcement learning (RL), where a learner is provided with an offline dataset from different tasks that share a common representation and is asked to learn the shared representation. We theoretically investigate offline multitask low-rank RL, and propose a new algorithm called MORL for offline multitask representation learning. Furthermore, we examine downstream RL in reward-free, offline and online scenarios, where a new task is introduced to the agent that shares the same representation as the upstream offline tasks. Our theoretical results demonstrate the benefits of using the learned representation from the upstream offline task instead of directly learning the representation of the low-rank model.

Learning from Observation (LfO) aims to imitate experts by learning from state-only demonstrations without requiring action labels. Existing adversarial imitation learning approaches learn a generator agent policy to produce state transitions that are indistinguishable to a discriminator that learns to classify agent and expert state transitions. Despite its simplicity in formulation, these methods are often sensitive to hyperparameters and brittle to train. Motivated by the recent success of diffusion models in generative modeling, we propose to integrate a diffusion model into the adversarial imitation learning from observation framework. Specifically, we employ a diffusion model to capture expert and agent transitions by generating the next state, given the current state. Then, we reformulate the learning objective to train the diffusion model as a binary classifier and use it to provide ``realness'' rewards for policy learning. Our proposed framework, Diffusion Imitation from Observation (DIFO), demonstrates superior performance in various continuous control domains, including navigation, locomotion, manipulation, and games.

Goal-conditioned reinforcement learning is a powerful way to control an AI agent's behavior at runtime. That said, popular goal representations, e.g., target states or natural language, are either limited to Markovian tasks or rely on ambiguous task semantics. We propose representing temporal goals using compositions of deterministic finite automata (cDFAs) and use cDFAs to guide RL agents. cDFAs balance the need for formal temporal semantics with ease of interpretation: if one can understand a flow chart, one can understand a cDFA. On the other hand, cDFAs form a countably infinite concept class with Boolean semantics, and subtle changes to the automaton can result in very different tasks, making them difficult to condition agent behavior on. To address this, we observe that all paths through a DFA correspond to a series of reach-avoid tasks and propose pre-training graph neural network embeddings on "reach-avoid derived" DFAs. Through empirical evaluation, we demonstrate that the proposed pre-training method enables zero-shot generalization to various cDFA task classes and accelerated policy specialization without the myopic suboptimality of hierarchical methods.

We propose a method of offline reinforcement learning (RL) featuring the performance guarantee without any assumptions on the data support. Under such conditions, estimating or optimizing the conventional performance metric is generally infeasible due to the distributional discrepancy between data and target policy distributions. To address this issue, we employ a worst-case policy value as a new metric and constructively show that the sample complexity bound of $O(\epsilon^{−2})$ is attainable without any data-support conditions, where $\epsilon>0$ is the policy suboptimality in the new metric. Moreover, as the new metric generalizes the conventional one, the algorithm can address standard offline RL tasks without modification. In this context, our sample complexity bound can be seen as a strict improvement on the previous bounds under the single-policy concentrability and the single-policy realizability.

Q-learning is widely used algorithm in reinforcement learning (RL) community. Under the lookup table setting, its convergence is well established. However, its behavior is known to be unstable with the linear function approximation case. This paper develops a new Q-learning algorithm, called RegQ, that converges when linear function approximation is used. We prove that simply adding an appropriate regularization term ensures convergence of the algorithm. Its stability is established using a recent analysis tool based on switching system models. Moreover, we experimentally show that RegQ converges in environments where Q-learning with linear function approximation has known to diverge. An error bound on the solution where the algorithm converges is also given.

Photo finishing tuning aims to automate the manual tuning process of the photo finishing pipeline, like Adobe Lightroom or Darktable. Previous works either use zeroth-order optimization, which is slow when the set of parameters increases, or rely on a differentiable proxy of the target finishing pipeline, which is hard to train. To overcome these challenges, we propose a novel goal-conditioned reinforcement learning framework for efficiently tuning parameters using a goal image as a condition. Unlike previous approaches, our tuning framework does not rely on any proxy and treats the photo finishing pipeline as a black box. Utilizing a trained reinforcement learning policy, it can efficiently find the desired set of parameters within just 10 queries, while optimization based approaches normally take 200 queries. Furthermore, our architecture utilizes a goal image to guide the iterative tuning of pipeline parameters, allowing for flexible conditioning on pixel-aligned target images, style images, or any other visually representable goals. We conduct detailed experiments on photo finishing tuning and photo stylization tuning tasks, demonstrating the advantages of our method.

The performance of modern reinforcement learning algorithms critically relies on tuning ever increasing numbers of hyperparameters. Often, small changes in a hyperparameter can lead to drastic changes in performance, and different environments require very different hyperparameter settings to achieve state-of-the-art performance reported in the literature. We currently lack a scalable and widely accepted approach to characterizing these complex interactions. This work proposes a new empirical methodology for studying, comparing, and quantifying the sensitivity of an algorithm’s performance to hyperparameter tuning for a given set of environments. We then demonstrate the utility of this methodology by assessing the hyperparameter sensitivity of several commonly used normalization variants of PPO. The results suggest that several algorithmic performance improvements may, in fact, be a result of an increased reliance on hyperparameter tuning.

A key source of complexity in next-generation AI models is the size of model outputs, making it time-consuming to parse and provide reliable feedback on. To ensure such models are aligned, we will need to bolster our understanding of scalable oversight and how to scale up human feedback. To this end, we study the challenges of scalable oversight in the context of goal-conditioned hierarchical reinforcement learning. Hierarchical structure is a promising entrypoint into studying how to scale up human feedback, which in this work we assume can only be provided for model outputs below a threshold size. In the cardinal feedback setting, we develop an apt sub-MDP reward and algorithm that allows us to acquire and scale up low-level feedback for learning with sublinear regret. In the ordinal feedback setting, we show the necessity of both high- and low-level feedback, and develop a hierarchical experimental design algorithm that efficiently acquires both types of feedback for learning. Altogether, our work aims to consolidate the foundations of scalable oversight, formalizing and studying the various challenges thereof.

This paper explores the realm of infinite horizon average reward Constrained Markov Decision Processes (CMDPs). To the best of our knowledge, this work is the first to delve into the regret and constraint violation analysis of average reward CMDPs with a general policy parametrization. To address this challenge, we propose a primal dual-based policy gradient algorithm that adeptly manages the constraints while ensuring a low regret guarantee toward achieving a global optimal policy. In particular, our proposed algorithm achieves $\tilde{\mathcal{O}}({T}^{4/5})$ objective regret and $\tilde{\mathcal{O}}({T}^{4/5})$ constraint violation bounds.

Offline reinforcement learning (RL) aims to train agents from pre-collected datasets. However, this comes with the added challenge of estimating the value of behaviors not covered in the dataset. Model-based methods offer a potential solution by training an approximate dynamics model, which then allows collection of additional synthetic data via rollouts in this model. The prevailing theory treats this approach as online RL in an approximate dynamics model, and any remaining performance gap is therefore understood as being due to dynamics model errors. In this paper, we analyze this assumption and investigate how popular algorithms perform as the learned dynamics model is improved. In contrast to both intuition and theory, if the learned dynamics model is replaced by the true error-free dynamics, existing model-based methods completely fail. This reveals a key oversight: The theoretical foundations assume sampling of full horizon rollouts in the learned dynamics model; however, in practice, the number of model-rollout steps is aggressively reduced to prevent accumulating errors. We show that this truncation of rollouts results in a set of edge-of-reach states at which we are effectively "bootstrapping from the void." This triggers pathological value overestimation and complete performance collapse. We term this the edge-of-reach problem. Based on this new insight, we fill important gaps in existing theory, and reveal how prior model-based methods are primarily addressing the edge-of-reach problem, rather than model-inaccuracy as claimed. Finally, we propose Reach-Aware Value Learning (RAVL), a simple and robust method that directly addresses the edge-of-reach problem and hence - unlike existing methods - does not fail as the dynamics model is improved. Since world models will inevitably improve, we believe this is a key step towards future-proofing offline RL.

*Constrained Reinforcement Learning* (CRL) tackles sequential decision-making problems where agents are required to achieve goals by maximizing the expected return while meeting domain-specific constraints, which are often formulated on expected costs. In this setting, *policy-based* methods are widely used since they come with several advantages when dealing with continuous-control problems. These methods search in the policy space with an *action-based* or *parameter-based* exploration strategy, depending on whether they learn directly the parameters of a stochastic policy or those of a stochastic hyperpolicy. In this paper, we propose a general framework for addressing CRL problems via *gradient-based primal-dual* algorithms, relying on an alternate ascent/descent scheme with dual-variable regularization. We introduce an exploration-agnostic algorithm, called C-PG, which exhibits global last-iterate convergence guarantees under (weak) gradient domination assumptions, improving and generalizing existing results. Then, we design C-PGAE and C-PGPE, the action-based and the parameter-based versions of C-PG, respectively, and we illustrate how they naturally extend to constraints defined in terms of *risk measures* over the costs, as it is often requested in safety-critical scenarios. Finally, we numerically validate our algorithms on constrained control problems, and compare them with state-of-the-art baselines, demonstrating their effectiveness.

Constraint-based offline reinforcement learning (RL) involves policy constraints or imposing penalties on the value function to mitigate overestimation errors caused by distributional shift. This paper focuses on a limitation in existing offline RL methods with penalized value function, indicating the potential for underestimation bias due to unnecessary bias introduced in the value function. To address this concern, we propose Exclusively Penalized Q-learning (EPQ), which reduces estimation bias in the value function by selectively penalizing states that are prone to inducing estimation errors. Numerical results show that our method significantly reduces underestimation bias and improves performance in various offline control tasks compared to other offline RL methods.

We consider a constrained Markov Decision Problem (CMDP) where the goal of an agent is to maximize the expected discounted sum of rewards over an infinite horizon while ensuring that the expected discounted sum of costs exceeds a certain threshold. Building on the idea of momentum-based acceleration, we develop the Primal-Dual Accelerated Natural Policy Gradient (PD-ANPG) algorithm that ensures an $\epsilon$ global optimality gap and $\epsilon$ constraint violation with $\tilde{\mathcal{O}}((1-\gamma)^{-7}\epsilon^{-2})$ sample complexity for general parameterized policies where $\gamma$ denotes the discount factor. This improves the state-of-the-art sample complexity in general parameterized CMDPs by a factor of $\mathcal{O}((1-\gamma)^{-1}\epsilon^{-2})$ and achieves the theoretical lower bound in $\epsilon^{-1}$.

To address the challenges of sim-to-real gap and sample efficiency in reinforcement learning (RL), this work studies distributionally robust Markov decision processes (RMDPs) --- optimize the worst-case performance when the deployed environment is within an uncertainty set around some nominal MDP. Despite recent efforts, the sample complexity of RMDPs has remained largely undetermined. While the statistical implications of distributional robustness in RL have been explored in some specific cases, the generalizability of the existing findings remains unclear, especially in comparison to standard RL. Assuming access to a generative model that samples from the nominal MDP, we examine the sample complexity of RMDPs using a class of generalized $L_p$ norms as the 'distance' function for the uncertainty set, under two commonly adopted $sa$-rectangular and $s$-rectangular conditions. Our results imply that RMDPs can be more sample-efficient to solve than standard MDPs using generalized $L_p$ norms in both $sa$- and $s$-rectangular cases, potentially inspiring more empirical research. We provide a near-optimal upper bound and a matching minimax lower bound for the $sa$-rectangular scenarios. For $s$-rectangular cases, we improve the state-of-the-art upper bound and also derive a lower bound using $L_\infty$ norm that verifies the tightness.