Reinforcement learning is an important type of Machine Learning where an agent learn how to behave in a environment by performing actions and seeing the results. In recent years, we've seen a lot of improvements in this fascinating area of research. Examples include DeepMind and the Deep Q learning architecture in 2014, beating the champion of the game of Go with AlphaGo in 2016, OpenAI and the PPO in 2017, amongst others. In this series of articles, we will focus on learning the different architectures used today to solve Reinforcement Learning problems. These will include Q -learning, Deep Q-learning, Policy Gradients, Actor Critic, and PPO.

This work adopts the very successful distributional perspective on reinforcement learning and adapts it to the continuous control setting. We combine this within a distributed framework for off-policy learning in order to develop what we call the Distributed Distributional Deep Deterministic Policy Gradient algorithm, D4PG. We also combine this technique with a number of additional, simple improvements such as the use of $N$-step returns and prioritized experience replay. Experimentally we examine the contribution of each of these individual components, and show how they interact, as well as their combined contributions. Our results show that across a wide variety of simple control tasks, difficult manipulation tasks, and a set of hard obstacle-based locomotion tasks the D4PG algorithm achieves state of the art performance.

In this report, we present a new reinforcement learning (RL) benchmark based on the Sonic the Hedgehog (TM) video game franchise. This benchmark is intended to measure the performance of transfer learning and few-shot learning algorithms in the RL domain. We also present and evaluate some baseline algorithms on the new benchmark.

A critical and challenging problem in reinforcement learning is how to learn the state-action value function from the experience replay buffer and simultaneously keep sample efficiency and faster convergence to a high quality solution. In prior works, transitions are uniformly sampled at random from the replay buffer or sampled based on their priority measured by temporal-difference (TD) error. However, these approaches do not fully take into consideration the intrinsic characteristics of transition distribution in the state space and could result in redundant and unnecessary TD updates, slowing down the convergence of the learning procedure. To overcome this problem, we propose a novel state distribution-aware sampling method to balance the replay times for transitions with skew distribution, which takes into account both the occurrence frequencies of transitions and the uncertainty of state-action values. Consequently, our approach could reduce the unnecessary TD updates and increase the TD updates for state-action value with more uncertainty, making the experience replay more effective and efficient. Extensive experiments are conducted on both classic control tasks and Atari 2600 games based on OpenAI gym platform and the experimental results demonstrate the effectiveness of our approach in comparison with the standard DQN approach.

Projective simulation (PS) is a model for intelligent agents with a deliberation capacity that is based on episodic memory. The model has been shown to provide a flexible framework for constructing reinforcement-learning agents, and it allows for quantum mechanical generalization, which leads to a speed-up in deliberation time. PS agents have been applied successfully in the context of complex skill learning in robotics, and in the design of state-of-the-art quantum experiments. In this paper, we study the performance of projective simulation in two benchmarking problems in navigation, namely the grid world and the mountain car problem. The performance of PS is compared to standard tabular reinforcement learning approaches, Q-learning and SARSA. Our comparison demonstrates that the performance of PS and standard learning approaches are qualitatively and quantitatively similar, while it is much easier to choose optimal model parameters in case of projective simulation, with a reduced computational effort of one to two orders of magnitude. Our results show that the projective simulation model stands out for its simplicity in terms of the number of model parameters, which makes it simple to set up the learning agent in unknown task environments.

Deep Reinforcement Learning (deep RL) has made several breakthroughs in recent years in applications ranging from complex control tasks in unmanned vehicles to game playing. Despite their success, deep RL still lacks several important capacities of human intelligence, such as transfer learning, abstraction and interpretability. Deep Symbolic Reinforcement Learning (DSRL) seeks to incorporate such capacities to deep Q-networks (DQN) by learning a relevant symbolic representation prior to using Q-learning. In this paper, we propose a novel extension of DSRL, which we call Symbolic Reinforcement Learning with Common Sense (SRL CS), offering a better balance between generalization and specialization, inspired by principles of common sense when assigning rewards and aggregating Q-values. Experiments reported in this paper show that SRL CS learns consistently faster than Q-learning and DSRL, achieving also a higher accuracy. In the hardest case, where agents were trained in a deterministic environment and tested in a random environment, SRL CS achieves nearly 100% average accuracy compared to DSRL's 70% and DQN's 50% accuracy. To the best of our knowledge, this is the first case of near perfect zero-shot transfer learning using Reinforcement Learning.

Multiple automakers have in development or in production automated driving systems (ADS) that offer freeway-pilot functions. This type of ADS is typically limited to restricted-access freeways only, that is, the transition from manual to automated modes takes place only after the ramp merging process is completed manually. One major challenge to extend the automation to ramp merging is that the automated vehicle needs to incorporate and optimize long-term objectives (e.g. successful and smooth merge) when near-term actions must be safely executed. Moreover, the merging process involves interactions with other vehicles whose behaviors are sometimes hard to predict but may influence the merging vehicle optimal actions. To tackle such a complicated control problem, we propose to apply Deep Reinforcement Learning (DRL) techniques for finding an optimal driving policy by maximizing the long-term reward in an interactive environment. Specifically, we apply a Long Short-Term Memory (LSTM) architecture to model the interactive environment, from which an internal state containing historical driving information is conveyed to a Deep Q-Network (DQN). The DQN is used to approximate the Q-function, which takes the internal state as input and generates Q-values as output for action selection. With this DRL architecture, the historical impact of interactive environment on the long-term reward can be captured and taken into account for deciding the optimal control policy. The proposed architecture has the potential to be extended and applied to other autonomous driving scenarios such as driving through a complex intersection or changing lanes under varying traffic flow conditions.

One of the most significant bottleneck in training large scale machine learning models on parameter server (PS) is the communication overhead, because it needs to frequently exchange the model gradients between the workers and servers during the training iterations. Gradient quantization has been proposed as an effective approach to reducing the communication volume. One key issue in gradient quantization is setting the number of bits for quantizing the gradients. Small number of bits can significantly reduce the communication overhead while hurts the gradient accuracies, and vise versa. An ideal quantization method would dynamically balance the communication overhead and model accuracy, through adjusting the number bits according to the knowledge learned from the immediate past training iterations. Existing methods, however, quantize the gradients either with fixed number of bits, or with predefined heuristic rules. In this paper we propose a novel adaptive quantization method within the framework of reinforcement learning. The method, referred to as MQGrad, formalizes the selection of quantization bits as actions in a Markov decision process (MDP) where the MDP states records the information collected from the past optimization iterations (e.g., the sequence of the loss function values). During the training iterations of a machine learning algorithm, MQGrad continuously updates the MDP state according to the changes of the loss function. Based on the information, MDP learns to select the optimal actions (number of bits) to quantize the gradients. Experimental results based on a benchmark dataset showed that MQGrad can accelerate the learning of a large scale deep neural network while keeping its prediction accuracies.

In this paper we discuss policy iteration methods for approximate solution of a finite-state discounted Markov decision problem, with a focus on feature-based aggregation methods and their connection with deep reinforcement learning schemes. We introduce features of the states of the original problem, and we formulate a smaller "aggregate" Markov decision problem, whose states relate to the features. The optimal cost function of the aggregate problem, a nonlinear function of the features, serves as an architecture for approximation in value space of the optimal cost function or the cost functions of policies of the original problem. We discuss properties and possible implementations of this type of aggregation, including a new approach to approximate policy iteration. In this approach the policy improvement operation combines feature-based aggregation with reinforcement learning based on deep neural networks, which is used to obtain the needed features. We argue that the cost function of a policy may be approximated much more accurately by the nonlinear function of the features provided by aggregation, than by the linear function of the features provided by deep reinforcement learning, thereby potentially leading to more effective policy improvement.

This post is a step by step guide through the paper. We'll cover the technical details and also walk through how you can get a version running on your own machine. Similarly to my post on AlphaZero, I'm not associated with the authors of the paper but just wanted to share my interpretation of their terrific work. We're going to build a reinforcement learning algorithm (an'agent') that gets good at driving a car around a 2D racetrack. At each time-step, the algorithm is fed an observation (a 64 x 64 pixel colour image of the car and immediate surroundings) and needs to return the next set of actions to take -- specifically, the steering direction (-1 to 1), acceleration (0 to 1) and brake (0 to 1). This action is then passed to the environment, which returns the next observation and the cycle starts again.