ICLR 2023 论文列表

We consider the sequential decision making problem of learning from an expert that has access to more information than the learner. For many problems this extra information will enable the expert to achieve greater long term reward than any policy without this privileged information access. We call these experts ``Impossibly Good'' because no learning algorithm will be able to reproduce their behavior. However, in these settings it is reasonable to attempt to recover the best policy possible given the agent's restricted access to information. We provide a set of necessary criteria on the expert that will allow a learner to recover the optimal policy in the reduced information space from the expert's advice alone. We also provide a new approach called Elf Distillation (Explorer Learning from Follower) that can be used in cases where these criteria are not met and environmental rewards must be taken into account. We show that this algorithm performs better than a variety of strong baselines on a challenging suite of Minigrid and Vizdoom environments.

We study reinforcement learning (RL) with linear function approximation, unknown transition, and adversarial losses in the bandit feedback setting. Specifically, the unknown transition probability function is a linear mixture model \citep{AyoubJSWY20,ZhouGS21,HeZG22} with a given feature mapping, and the learner only observes the losses of the experienced state-action pairs instead of the whole loss function. We propose an efficient algorithm LSUOB-REPS which achieves $\widetilde{O}(dS^2\sqrt{K}+\sqrt{HSAK})$ regret guarantee with high probability, where $d$ is the ambient dimension of the feature mapping, $S$ is the size of the state space, $A$ is the size of the action space, $H$ is the episode length and $K$ is the number of episodes. Furthermore, we also prove a lower bound of order $\Omega(dH\sqrt{K}+\sqrt{HSAK})$ for this setting. To the best of our knowledge, we make the first step to establish a provably efficient algorithm with a sublinear regret guarantee in this challenging setting and solve the open problem of \citet{HeZG22}.

Deep reinforcement learning~(RL) has achieved remarkable successes in complex single-task settings. However, designing RL agents that can learn multiple tasks and leverage prior experience to quickly adapt to a related new task remains challenging. Despite previous attempts to improve on these areas, our understanding of multi-task training and generalization in RL remains limited. To fill this gap, we investigate the generalization capabilities of a popular actor-critic method, IMPALA. Specifically, we build on previous work that has advocated for the use of modes and difficulties of Atari 2600 games as a challenging benchmark for transfer learning in RL. We do so by pretraining an agent on multiple variants of the same Atari game before fine-tuning on the remaining never-before-seen variants. This protocol simplifies the multi-task pretraining phase by limiting negative interference between tasks and allows us to better understand the dynamics of multi-task training and generalization. We find that, given a fixed amount of pretraining data, agents trained with more variations are able to generalize better. Surprisingly, we also observe that this advantage can still be present after fine-tuning for 200M environment frames than when doing zero-shot transfer. This highlights the potential effect of a good learned representation. We also find that, even though small networks have remained popular to solve Atari 2600 games, increasing the capacity of the value and policy network is critical to achieve good performance as we increase the number of pretraining modes and difficulties. Overall, our findings emphasize key points that are essential for efficient multi-task training and generalization in reinforcement learning.

Reinforcement Learning (RL) agents are often unable to generalise well to environment variations in the state space that were not observed during training. This issue is especially problematic for image-based RL, where a change in just one variable, such as the background colour, can change many pixels in the image. The changed pixels can lead to drastic changes in the agent's latent representation of the image, causing the learned policy to fail. To learn more robust representations, we introduce TEmporal Disentanglement (TED), a self-supervised auxiliary task that leads to disentangled image representations exploiting the sequential nature of RL observations. We find empirically that RL algorithms utilising TED as an auxiliary task adapt more quickly to changes in environment variables with continued training compared to state-of-the-art representation learning methods. Since TED enforces a disentangled structure of the representation, our experiments also show that policies trained with TED generalise better to unseen values of variables irrelevant to the task (e.g. background colour) as well as unseen values of variables that affect the optimal policy (e.g. goal positions).

Recent improvements in conditional generative modeling have made it possible to generate high-quality images from language descriptions alone. We investigate whether these methods can directly address the problem of sequential decision-making. We view decision-making not through the lens of reinforcement learning (RL), but rather through conditional generative modeling. To our surprise, we find that our formulation leads to policies that can outperform existing offline RL approaches across standard benchmarks. By modeling a policy as a return-conditional generative model, we avoid the need for dynamic programming and subsequently eliminate many of the complexities that come with traditional offline RL. We further demonstrate the advantages of modeling policies as conditional generative models by considering two other conditioning variables: constraints and skills. Conditioning on a single constraint or skill during training leads to behaviors at test-time that can satisfy several constraints together or demonstrate a composition of skills. Our results illustrate that conditional generative modeling is a powerful tool for decision-making.

Many important tasks involve some notion of long-term progress in multiple phases: e.g. to clean a shelf it must be cleared of items, cleaning products applied, and then the items placed back on the shelf. In this work, we explore the use of expert demonstrations in long-horizon tasks to learn a monotonically increasing function that summarizes progress. This function can then be used to aid agent exploration in environments with sparse rewards. As a case study we consider the NetHack environment, which requires long-term progress at a variety of scales and is far from being solved by existing approaches. In this environment, we demonstrate that by learning a model of long-term progress from expert data containing only observations, we can achieve efficient exploration in challenging sparse tasks, well beyond what is possible with current state-of-the-art approaches. We have made the curated gameplay dataset used in this work available at https://github.com/deepmind/nao_top10.

Open-ended learning methods that automatically generate a curriculum of increasingly challenging tasks serve as a promising avenue toward generally capable reinforcement learning agents. Existing methods adapt curricula independently over either environment parameters (in single-agent settings) or co-player policies (in multi-agent settings). However, the strengths and weaknesses of co-players can manifest themselves differently depending on environmental features. It is thus crucial to consider the dependency between the environment and co-player when shaping a curriculum in multi-agent domains. In this work, we use this insight and extend Unsupervised Environment Design (UED) to multi-agent environments. We then introduce Multi-Agent Environment Design Strategist for Open-Ended Learning (MAESTRO), the first multi-agent UED approach for two-player zero-sum settings. MAESTRO efficiently produces adversarial, joint curricula over both environments and co-players and attains minimax-regret guarantees at Nash equilibrium. Our experiments show that MAESTRO outperforms a number of strong baselines on competitive two-player games, spanning discrete and continuous control settings.

Despite having achieved spectacular milestones in an array of important real-world applications, most Reinforcement Learning (RL) methods are very brittle concerning their hyperparameters. Notwithstanding the crucial importance of setting the hyperparameters in training state-of-the-art agents, the task of hyperparameter optimization (HPO) in RL is understudied. In this paper, we propose a novel gray-box Bayesian Optimization technique for HPO in RL, that enriches Gaussian Processes with reward curve estimations based on generalized logistic functions. In a very large-scale experimental protocol, comprising 5 popular RL methods (DDPG, A2C, PPO, SAC, TD3), dozens of environments (Atari, Mujoco), and 7 HPO baselines, we demonstrate that our method significantly outperforms current HPO practices in RL.

Co-design of morphology and control of a Voxel-based Soft Robot (VSR) is challenging due to the notorious bi-level optimization. In this paper, we present a Curriculum-based Co-design (CuCo) method for learning to design and control VSRs through an easy-to-difficult process. Specifically, we expand the design space from a small size to the target size gradually through a predefined curriculum. At each learning stage of the curriculum, we use reinforcement learning to simultaneously train the design policy and the control policy, which is enabled by incorporating the design process into the environment and using differentiable policy representations. The converged morphology and the learned policies from last stage are inherited and then serve as the starting point for the next stage. In empirical studies, we show that CuCo is more efficient in creating larger robots with better performance by reusing the practical design and control patterns learned within each stage, in comparison to prior approaches that learn from scratch in the space of target size.

By properly utilizing the learned environment model, model-based reinforcement learning methods can improve the sample efficiency for decision-making problems. Beyond using the learned environment model to train a policy, the success of MCTS-based methods shows that directly incorporating the learned environment model as a planner to make decisions might be more effective. However, when action space is of high dimension and continuous, directly planning according to the learned model is costly and non-trivial. Because of two challenges: (1) the infinite number of candidate actions and (2) the temporal dependency between actions in different timesteps. To address these challenges, inspired by Differential Dynamic Programming (DDP) in optimal control theory, we design a novel Policy Optimization with Model Planning (POMP) algorithm, which incorporates a carefully designed Deep Differential Dynamic Programming (D3P) planner into the model-based RL framework. In D3P planner, (1) to effectively plan in the continuous action space, we construct a locally quadratic programming problem that uses a gradient-based optimization process to replace search. (2) To take the temporal dependency of actions at different timesteps into account, we leverage the updated and latest actions of previous timesteps (i.e., step $1, \cdots, h-1$) to update the action of the current step (i.e., step $h$), instead of updating all actions simultaneously. We theoretically prove the convergence rate for our D3P planner and analyze the effect of the feedback term. In practice, to effectively apply the neural network based D3P planner in reinforcement learning, we leverage the policy network to initialize the action sequence and keep the action update conservative in the planning process. Experiments demonstrate that POMP consistently improves sample efficiency on widely used continuous control tasks. Our code is released at https://github.com/POMP-D3P/POMP-D3P.

MOBA games, e.g., Dota2 and Honor of Kings, have been actively used as the testbed for the recent AI research on games, and various AI systems have been developed at the human level so far. However, these AI systems mainly focus on how to compete with humans, less on exploring how to collaborate with humans. To this end, this paper makes the first attempt to investigate human-agent collaboration in MOBA games. In this paper, we propose to enable humans and agents to collaborate through explicit communication by designing an efficient and interpretable Meta-Command Communication-based framework, dubbed MCC, for accomplishing effective human-agent collaboration in MOBA games. The MCC framework consists of two pivotal modules: 1) an interpretable communication protocol, i.e., the Meta-Command, to bridge the communication gap between humans and agents; 2) a meta-command value estimator, i.e., the Meta-Command Selector, to select a valuable meta-command for each agent to achieve effective human-agent collaboration. Experimental results in Honor of Kings demonstrate that MCC agents can collaborate reasonably well with human teammates and even generalize to collaborate with different levels and numbers of human teammates. Videos are available at https://sites.google.com/view/mcc-demo.

We introduce our Maximum-Entropy Rewarded Reinforcement Learning (MERRL) framework that selects training data for more accurate Natural Language Processing (NLP). Because conventional data selection methods select training samples based on the test domain knowledge and not on real life data, they frequently fail in unknown domains like patent and Twitter. Our approach selects training samples that maximize information uncertainty measured by entropy, including observation entropy like empirical Shannon entropy, Min-entropy, R\'enyi entropy, and prediction entropy using mutual information, to cover more possible queries that may appear in unknown worlds. Our MERRL using regularized A2C and SAC achieves up to -99.7 perplexity decrease (-43.4\% relatively) in language modeling, +25.0 accuracy increase (+40.0\% relatively) in sentiment analysis, and +5.0 F1 score increase (+30.8\% relatively) in named entity recognition over various domains, demonstrating strong generalization power on unknown test sets.

We propose a novel framework for multitask reinforcement learning based on the minimum description length (MDL) principle. In this approach, which we term MDL-control (MDL-C), the agent learns the common structure among the tasks with which it is faced and then distills it into a simpler representation which facilitates faster convergence and generalization to new tasks. In doing so, MDL-C naturally balances adaptation to each task with epistemic uncertainty about the task distribution. We motivate MDL-C via formal connections between the MDL principle and Bayesian inference, derive theoretical performance guarantees, and demonstrate MDL-C's empirical effectiveness on both discrete and high-dimensional continuous control tasks.

Past research on interactive decision making problems (bandits, reinforcement learning, etc.) mostly focuses on the minimax regret that measures the algorithm's performance on the hardest instance. However, an ideal algorithm should adapt to the complexity of a particular problem instance and incur smaller regrets on easy instances than worst-case instances. In this paper, we design the first asymptotic instance-optimal algorithm for general interactive decision making problems with finite number of decisions under mild conditions. On every instance $f$, our algorithm outperforms all consistent algorithms (those achieving non-trivial regrets on all instances), and has asymptotic regret $\mathcal{C}(f) \ln n$, where $\mathcal{C}(f)$ is an exact characterization of the complexity of $f$. The key step of the algorithm involves hypothesis testing with active data collection. It computes the most economical decisions with which the algorithm collects observations to test whether an estimated instance is indeed correct; thus, the complexity $\mathcal{C}(f)$ is the minimum cost to test the instance $f$ against other instances. Our results, instantiated on concrete problems, recover the classical gap-dependent bounds for multi-armed bandits and prior works on linear bandits, and improve upon the previous best instance-dependent upper bound for reinforcement learning.

Auxiliary tasks improve the representations learned by deep reinforcement learning agents. Analytically, their effect is reasonably well-understood; in practice, how-ever, their primary use remains in support of a main learning objective, rather than as a method for learning representations. This is perhaps surprising given that many auxiliary tasks are defined procedurally, and hence can be treated as an essentially infinite source of information about the environment. Based on this observation, we study the effectiveness of auxiliary tasks for learning rich representations, focusing on the setting where the number of tasks and the size of the agent’s network are simultaneously increased. For this purpose, we derive a new family of auxiliary tasks based on the successor measure. These tasks are easy to implement and have appealing theoretical properties. Combined with a suitable off-policy learning rule, the result is a representation learning algorithm that can be understood as extending Mahadevan & Maggioni (2007)’s proto-value functions to deep reinforcement learning – accordingly, we call the resulting object proto-value networks. Through a series of experiments on the Arcade Learning Environment, we demonstrate that proto-value networks produce rich features that may be used to obtain performance comparable to established algorithms, using only linear approximation and a small number (~4M) of interactions with the environment’s reward function.

Symbolic regression (SR) aims to discover concise closed-form mathematical equations from data, a task fundamental to scientific discovery. However, the problem is highly challenging because closed-form equations lie in a complex combinatorial search space. Existing methods, ranging from heuristic search to reinforcement learning, fail to scale with the number of input variables. We make the observation that closed-form equations often have structural characteristics and invariances (e.g. the commutative law) that could be further exploited to build more effective symbolic regression solutions. Motivated by this observation, our key contribution is to leverage pre-trained deep generative models to capture the intrinsic regularities of equations, thereby providing a solid foundation for subsequent optimization steps. We show that our novel formalism unifies several prominent approaches of symbolic regression and offers a new perspective to justify and improve on the previous ad hoc designs, such as the usage of cross-entropy loss during pre-training. Specifically, we propose an instantiation of our framework, Deep Generative Symbolic Regression (DGSR). In our experiments, we show that DGSR achieves a higher recovery rate of true equations in the setting of a larger number of input variables, and it is more computationally efficient at inference time than state-of-the-art RL symbolic regression solutions.

Recent methods for imitation learning directly learn a $Q$-function using an implicit reward formulation rather than an explicit reward function. However, these methods generally require implicit reward regularization to improve stability and often mistreat absorbing states. Previous works show that a squared norm regularization on the implicit reward function is effective, but do not provide a theoretical analysis of the resulting properties of the algorithms. In this work, we show that using this regularizer under a mixture distribution of the policy and the expert provides a particularly illuminating perspective: the original objective can be understood as squared Bellman error minimization, and the corresponding optimization problem minimizes a bounded $\chi^2$-Divergence between the expert and the mixture distribution. This perspective allows us to address instabilities and properly treat absorbing states. We show that our method, Least Squares Inverse Q-Learning (LS-IQ), outperforms state-of-the-art algorithms, particularly in environments with absorbing states. Finally, we propose to use an inverse dynamics model to learn from observations only. Using this approach, we retain performance in settings where no expert actions are available.

Zeroth-order (ZO) optimization, in which the derivative is unavailable, has recently succeeded in many important machine learning applications. Existing algorithms rely on finite difference (FD) methods for derivative estimation and gradient descent (GD)-based approaches for optimization. However, these algorithms suffer from query inefficiency because many additional function queries are required for derivative estimation in their every GD update, which typically hinders their deployment in real-world applications where every function query is expensive. To this end, we propose a trajectory-informed derivative estimation method which only employs the optimization trajectory (i.e., the history of function queries during optimization) and hence can eliminate the need for additional function queries to estimate a derivative. Moreover, based on our derivative estimation, we propose the technique of dynamic virtual updates, which allows us to reliably perform multiple steps of GD updates without reapplying derivative estimation. Based on these two contributions, we introduce the zeroth-order optimization with trajectory-informed derivative estimation (ZoRD) algorithm for query-efficient ZO optimization. We theoretically demonstrate that our trajectory-informed derivative estimation and our ZoRD algorithm improve over existing approaches, which is then supported by our real-world experiments such as black-box adversarial attack, non-differentiable metric optimization, and derivative-free reinforcement learning.

Partial Observability---where agents can only observe partial information about the true underlying state of the system---is ubiquitous in real-world applications of Reinforcement Learning (RL). Theoretically, learning a near-optimal policy under partial observability is known to be hard in the worst case due to an exponential sample complexity lower bound. Recent work has identified several tractable subclasses that are learnable with polynomial samples, such as Partially Observable Markov Decision Processes (POMDPs) with certain revealing or decodability conditions. However, this line of research is still in its infancy, where (1) unified structural conditions enabling sample-efficient learning are lacking; (2) existing sample complexities for known tractable subclasses are far from sharp; and (3) fewer sample-efficient algorithms are available than in fully observable RL. This paper advances all three aspects above for Partially Observable RL in the general setting of Predictive State Representations (PSRs). First, we propose a natural and unified structural condition for PSRs called \emph{B-stability}. B-stable PSRs encompasses the vast majority of known tractable subclasses such as weakly revealing POMDPs, low-rank future-sufficient POMDPs, decodable POMDPs, and regular PSRs. Next, we show that any B-stable PSR can be learned with polynomial samples in relevant problem parameters. When instantiated in the aforementioned subclasses, our sample complexities improve substantially over the current best ones. Finally, our results are achieved by three algorithms simultaneously: Optimistic Maximum Likelihood Estimation, Estimation-to-Decisions, and Model-Based Optimistic Posterior Sampling. The latter two algorithms are new for sample-efficient learning of POMDPs/PSRs. We additionally design a variant of the Estimation-to-Decisions algorithm to perform sample-efficient \emph{all-policy model estimation} for B-stable PSRs, which also yields guarantees for reward-free learning as an implication.