We present a machine learning framework for multi-agent systems to learn both the optimal policy for maximizing the rewards and the encoding of the high dimensional visual observation. The encoding is useful for sharing local visual observations with other agents under communication resource constraints. The actor-encoder encodes the raw images and chooses an action based on local observations and messages sent by the other agents. The machine learning agent generates not only an actuator command to the physical device, but also a communication message to the other agents. We formulate a reinforcement learning problem, which extends the action space to consider the communication action as well. The feasibility of the reinforcement learning framework is demonstrated using a 3D simulation environment with two collaborating agents. The environment provides realistic visual observations to be used and shared between the two agents.

We introduce and analyse two algorithms for exploration-exploitation in discrete and continuous Markov Decision Processes (MDPs) based on exploration bonuses. SCAL$^+$ is a variant of SCAL (Fruit et al., 2018) that performs efficient exploration-exploitation in any unknown weakly-communicating MDP for which an upper bound C on the span of the optimal bias function is known. For an MDP with $S$ states, $A$ actions and $\Gamma \leq S$ possible next states, we prove that SCAL$^+$ achieves the same theoretical guarantees as SCAL (i.e., a high probability regret bound of $\widetilde{O}(C\sqrt{\Gamma SAT})$), with a much smaller computational complexity. Similarly, C-SCAL$^+$ exploits an exploration bonus to achieve sublinear regret in any undiscounted MDP with continuous state space. We show that C-SCAL$^+$ achieves the same regret bound as UCCRL (Ortner and Ryabko, 2012) while being the first implementable algorithm with regret guarantees in this setting. While optimistic algorithms such as UCRL, SCAL or UCCRL maintain a high-confidence set of plausible MDPs around the true unknown MDP, SCAL$^+$ and C-SCAL$^+$ leverage on an exploration bonus to directly plan on the empirically estimated MDP, thus being more computationally efficient.

Efficient Reinforcement Learning usually takes advantage of demonstration or good exploration strategy. By applying posterior sampling in model-free RL under the hypothesis of GP, we propose Gaussian Process Posterior Sampling Reinforcement Learning(GPPSTD) algorithm in continuous state space, giving theoretical justifications and empirical results. We also provide theoretical and empirical results that various demonstration could lower expected uncertainty and benefit posterior sampling exploration. In this way, we combined the demonstration and exploration process together to achieve a more efficient reinforcement learning.

Deep reinforcement-learning methods have achieved remarkable performance on challenging control tasks. Observations of the resulting behavior give the impression that the agent has constructed a generalized representation that supports insightful action decisions. We re-examine what is meant by generalization in RL, and propose several definitions based on an agent's performance in on-policy, off-policy, and unreachable states. We propose a set of practical methods for evaluating agents with these definitions of generalization. We demonstrate these techniques on a common benchmark task for deep RL, and we show that the learned networks make poor decisions for states that differ only slightly from on-policy states, even though those states are not selected adversarially. Taken together, these results call into question the extent to which deep Q-networks learn generalized representations, and suggest that more experimentation and analysis is necessary before claims of representation learning can be supported.

A key challenge for gradient based optimization methods in model-free reinforcement learning is to develop an approach that is sample efficient and has low variance. In this work, we apply Kronecker-factored curvature estimation technique (KFAC) to a recently proposed gradient estimator for control variate optimization, RELAX, to increase the sample efficiency of using this gradient estimation method in reinforcement learning. The performance of the proposed method is demonstrated on a synthetic problem and a set of three discrete control task Atari games.

It is a widely accepted principle that software without tests has bugs. Testing reinforcement learning agents is especially difficult because of the stochastic nature of both agents and environments, the complexity of state-of-the-art models, and the sequential nature of their predictions. Recently, the Arcade Learning Environment (ALE) has become one of the most widely used benchmark suites for deep learning research, and state-of-the-art Reinforcement Learning (RL) agents have been shown to routinely equal or exceed human performance on many ALE tasks. Since ALE is based on emulation of original Atari games, the environment does not provide semantically meaningful representations of internal game state. This means that ALE has limited utility as an environment for supporting testing or model introspection. We propose TOYBOX, a collection of reimplementations of these games that solves this critical problem and enables robust testing of RL agents.

The effectiveness of model-based versus model-free methods is a long-standing question in reinforcement learning (RL). Motivated by recent empirical success of RL on continuous control tasks, we study the sample complexity of popular model-based and model-free algorithms on the Linear Quadratic Regulator (LQR). We show that for policy evaluation, a simple model-based plugin method requires asymptotically less samples than the classical least-squares temporal difference (LSTD) estimator to reach the same quality of solution; the sample complexity gap between the two methods can be at least a factor of state dimension. For policy evaluation, we study a simple family of problem instances and show that nominal (certainty equivalence principle) control also requires a factor of state dimension fewer samples than the policy gradient method to reach the same level of control performance on these instances. Furthermore, the gap persists even when employing baselines commonly used in practice. To the best of our knowledge, this is the first theoretical result which demonstrates a separation in the sample complexity between model-based and model-free methods on a continuous control task.

The performance of adversarial dialogue generation models relies on the quality of the reward signal produced by the discriminator. The reward signal from a poor discriminator can be very sparse and unstable, which may lead the generator to fall into a local optimum or to produce nonsense replies. To alleviate the first problem, we first extend a recently proposed adversarial dialogue generation method to an adversarial imitation learning solution. Then, in the framework of adversarial inverse reinforcement learning, we propose a new reward model for dialogue generation that can provide a more accurate and precise reward signal for generator training. We evaluate the performance of the resulting model with automatic metrics and human evaluations in two annotation settings. Our experimental results demonstrate that our model can generate more high-quality responses and achieve higher overall performance than the state-of-the-art.

Importance weighting is a key ingredient in many algorithms for causal inference and related problems, such as off-policy evaluation for reinforcement learning. Recently, theorists proved that on separable data, unregularized linear networks, trained with cross-entropy loss and optimized by stochastic gradient descent converge in direction to the max margin solution. This solution depends on the location of data points but not their weights, nullifying the effect of importance weighting. This paper asks, for realistic deep networks, for which all datasets are separable, what is the effect of importance weighting? Lacking theoretical tools for analyzing modern deep (nonlinear, unregularized) networks, we investigate the question empirically on both realistic and synthetic data. Our results demonstrate that while importance weighting alters the learned model early in training, its effect diminishes to negligible with indefinite training. However, this diminishing effect does not occur in the presence of L2-regularization. These results (i) support the broader applicability of theoretical findings by Soudry et al (2018), who analyze linear networks; (ii) call into question the practice of importance weighting; and (iii) suggest that its usefulness interacts strongly with the early stopping criteria and regularization methods that interact with the loss function.

We propose a new method for learning from a single demonstration to solve hard exploration tasks like the Atari game Montezuma's Revenge. Instead of imitating human demonstrations, as proposed in other recent works, our approach is to maximize rewards directly. Our agent is trained using off-the-shelf reinforcement learning, but starts every episode by resetting to a state from a demonstration. By starting from such demonstration states, the agent requires much less exploration to learn a game compared to when it starts from the beginning of the game at every episode. We analyze reinforcement learning for tasks with sparse rewards in a simple toy environment, where we show that the run-time of standard RL methods scales exponentially in the number of states between rewards. Our method reduces this to quadratic scaling, opening up many tasks that were previously infeasible. We then apply our method to Montezuma's Revenge, for which we present a trained agent achieving a high-score of 74,500, better than any previously published result.