Reinforcement learning is time-consuming for complex tasks due to the need for large amounts of training data. Recent advances in GPU-based simulation, such as Isaac Gym, have sped up data collection thousands of times on a commodity GPU. Most prior works used on-policy methods like PPO due to their simplicity and ease of scaling. Off-policy methods are more data efficient but challenging to scale, resulting in a longer wall-clock training time. This paper presents a Parallel $Q$-Learning (PQL) scheme that outperforms PPO in wall-clock time while maintaining superior sample efficiency of off-policy learning. PQL achieves this by parallelizing data collection, policy learning, and value learning. Different from prior works on distributed off-policy learning, such as Apex, our scheme is designed specifically for massively parallel GPU-based simulation and optimized to work on a single workstation. In experiments, we demonstrate that $Q$-learning can be scaled to \textit{tens of thousands of parallel environments} and investigate important factors affecting learning speed. The code is available at https://github.com/Improbable-AI/pql.

As with any machine learning problem with limited data, effective offline RL algorithms require careful regularization to avoid overfitting. One-step methods perform regularization by doing just a single step of policy improvement, while critic regularization methods do many steps of policy improvement with a regularized objective. These methods appear distinct. One-step methods, such as advantage-weighted regression and conditional behavioral cloning, truncate policy iteration after just one step. This ``early stopping'' makes one-step RL simple and stable, but can limit its asymptotic performance. Critic regularization typically requires more compute but has appealing lower-bound guarantees. In this paper, we draw a close connection between these methods: applying a multi-step critic regularization method with a regularization coefficient of 1 yields the same policy as one-step RL. While practical implementations violate our assumptions and critic regularization is typically applied with smaller regularization coefficients, our experiments nevertheless show that our analysis makes accurate, testable predictions about practical offline RL methods (CQL and one-step RL) with commonly-used hyperparameters. Our results that every problem can be solved with a single step of policy improvement, but rather that one-step RL might be competitive with critic regularization on RL problems that demand strong regularization.

Punctuation restoration is an important task in automatic speech recognition (ASR) which aim to restore the syntactic structure of generated ASR texts to improve readability. While punctuated texts are abundant from written documents, the discrepancy between written punctuated texts and ASR texts limits the usability of written texts in training punctuation restoration systems for ASR texts. This paper proposes a reinforcement learning method to exploit in-topic written texts and recent advances in large pre-trained generative language models to bridge this gap. The experiments show that our method achieves state-of-the-art performance on the ASR test set on two benchmark datasets for punctuation restoration.

Model-based reinforcement learning (RL) has demonstrated remarkable successes on a range of continuous control tasks due to its high sample efficiency. To save the computation cost of conducting planning online, recent practices tend to distill optimized action sequences into an RL policy during the training phase. Although the distillation can incorporate both the foresight of planning and the exploration ability of RL policies, the theoretical understanding of these methods is yet unclear. In this paper, we extend the policy improvement step of Soft Actor-Critic (SAC) by developing an approach to distill from model-based planning to the policy. We then demonstrate that such an approach of policy improvement has a theoretical guarantee of monotonic improvement and convergence to the maximum value defined in SAC. We discuss effective design choices and implement our theory as a practical algorithm -- Model-based Planning Distilled to Policy (MPDP) -- that updates the policy jointly over multiple future time steps. Extensive experiments show that MPDP achieves better sample efficiency and asymptotic performance than both model-free and model-based planning algorithms on six continuous control benchmark tasks in MuJoCo.

We consider the problem of contextual bandits and imitation learning, where the learner lacks direct knowledge of the executed action's reward. Instead, the learner can actively query an expert at each round to compare two actions and receive noisy preference feedback. The learner's objective is two-fold: to minimize the regret associated with the executed actions, while simultaneously, minimizing the number of comparison queries made to the expert. In this paper, we assume that the learner has access to a function class that can represent the expert's preference model under appropriate link functions, and provide an algorithm that leverages an online regression oracle with respect to this function class for choosing its actions and deciding when to query. For the contextual bandit setting, our algorithm achieves a regret bound that combines the best of both worlds, scaling as $O(\min\{\sqrt{T}, d/\Delta\})$, where $T$ represents the number of interactions, $d$ represents the eluder dimension of the function class, and $\Delta$ represents the minimum preference of the optimal action over any suboptimal action under all contexts. Our algorithm does not require the knowledge of $\Delta$, and the obtained regret bound is comparable to what can be achieved in the standard contextual bandits setting where the learner observes reward signals at each round. Additionally, our algorithm makes only $O(\min\{T, d^2/\Delta^2\})$ queries to the expert. We then extend our algorithm to the imitation learning setting, where the learning agent engages with an unknown environment in episodes of length $H$ each, and provide similar guarantees for regret and query complexity. Interestingly, our algorithm for imitation learning can even learn to outperform the underlying expert, when it is suboptimal, highlighting a practical benefit of preference-based feedback in imitation learning.

The legitimacy of bottom-up democratic processes for the distribution of public funds by policy-makers is challenging and complex. Participatory budgeting is such a process, where voting outcomes may not always be fair or inclusive. Deliberation for which project ideas to put for voting and choose for implementation lack systematization and do not scale. This paper addresses these grand challenges by introducing a novel and legitimate iterative consensus-based participatory budgeting process. Consensus is designed to be a result of decision support via an innovative multi-agent reinforcement learning approach. Voters are assisted to interact with each other to make viable compromises. Extensive experimental evaluation with real-world participatory budgeting data from Poland reveal striking findings: Consensus is reachable, efficient and robust. Compromise is required, which is though comparable to the one of existing voting aggregation methods that promote fairness and inclusion without though attaining consensus.

The 2022 Russian invasion of Ukraine was accompanied by a large-scale, pro-Russian propaganda campaign on social media. However, the strategy behind the dissemination of propaganda has remained unclear, particularly how the online discourse was strategically shaped by the propagandists' community. Here, we analyze the strategy of the Twitter community using an inverse reinforcement learning (IRL) approach. Specifically, IRL allows us to model online behavior as a Markov decision process, where the goal is to infer the underlying reward structure that guides propagandists when interacting with users with a supporting or opposing stance toward the invasion. Thereby, we aim to understand empirically whether and how between-user interactions are strategically used to promote the proliferation of Russian propaganda. For this, we leverage a large-scale dataset with 349,455 posts with pro-Russian propaganda from 132,131 users. We show that bots and humans follow a different strategy: bots respond predominantly to pro-invasion messages, suggesting that they seek to drive virality; while messages indicating opposition primarily elicit responses from humans, suggesting that they tend to engage in critical discussions. To the best of our knowledge, this is the first study analyzing the strategy behind propaganda from the 2022 Russian invasion of Ukraine through the lens of IRL.

High-frequency displays are gaining immense popularity because of their increasing use in video games and virtual reality applications. However, the issue is that the underlying GPUs cannot continuously generate frames at this high rate -- this results in a less smooth and responsive experience. Furthermore, if the frame rate is not synchronized with the refresh rate, the user may experience screen tearing and stuttering. Previous works propose increasing the frame rate to provide a smooth experience on modern displays by predicting new frames based on past or future frames. Interpolation and extrapolation are two widely used algorithms that predict new frames. Interpolation requires waiting for the future frame to make a prediction, which adds additional latency. On the other hand, extrapolation provides a better quality of experience because it relies solely on past frames -- it does not incur any additional latency. The simplest method to extrapolate a frame is to warp the previous frame using motion vectors; however, the warped frame may contain improperly rendered visual artifacts due to dynamic objects -- this makes it very challenging to design such a scheme. Past work has used DNNs to get good accuracy, however, these approaches are slow. This paper proposes Exwarp -- an approach based on reinforcement learning (RL) to intelligently choose between the slower DNN-based extrapolation and faster warping-based methods to increase the frame rate by 4x with an almost negligible reduction in the perceived image quality.

Reinforcement learning (RL) has been shown to learn sophisticated control policies for complex tasks including games, robotics, heating and cooling systems and text generation. The action-perception cycle in RL, however, generally assumes that a measurement of the state of the environment is available at each time step without a cost. In applications such as materials design, deep-sea and planetary robot exploration and medicine, however, there can be a high cost associated with measuring, or even approximating, the state of the environment. In this paper, we survey the recently growing literature that adopts the perspective that an RL agent might not need, or even want, a costly measurement at each time step. Within this context, we propose the Deep Dynamic Multi-Step Observationless Agent (DMSOA), contrast it with the literature and empirically evaluate it on OpenAI gym and Atari Pong environments. Our results, show that DMSOA learns a better policy with fewer decision steps and measurements than the considered alternative from the literature.

Viewpoint planning is an important task in any application where objects or scenes need to be viewed from different angles to achieve sufficient coverage. The mapping of confined spaces such as shelves is an especially challenging task since objects occlude each other and the scene can only be observed from the front, posing limitations on the possible viewpoints. In this paper, we propose a deep reinforcement learning framework that generates promising views aiming at reducing the map entropy. Additionally, the pipeline extends standard viewpoint planning by predicting adequate minimally invasive push actions to uncover occluded objects and increase the visible space. Using a 2.5D occupancy height map as state representation that can be efficiently updated, our system decides whether to plan a new viewpoint or perform a push. To learn feasible pushes, we use a neural network to sample push candidates on the map based on training data provided by human experts. As simulated and real-world experimental results with a robotic arm show, our system is able to significantly increase the mapped space compared to different baselines, while the executed push actions highly benefit the viewpoint planner with only minor changes to the object configuration.