Models, code, and papers for "Nicolas Heess":
We present an off-policy actor-critic algorithm for Reinforcement Learning (RL) that combines ideas from gradient-free optimization via stochastic search with learned action-value function. The result is a simple procedure consisting of three steps: i) policy evaluation by estimating a parametric action-value function; ii) policy improvement via the estimation of a local non-parametric policy; and iii) generalization by fitting a parametric policy. Each step can be implemented in different ways, giving rise to several algorithm variants. Our algorithm draws on connections to existing literature on black-box optimization and 'RL as an inference' and it can be seen either as an extension of the Maximum a Posteriori Policy Optimisation algorithm (MPO) [Abdolmaleki et al., 2018a], or as an extension of Trust Region Covariance Matrix Adaptation Evolutionary Strategy (CMA-ES) [Abdolmaleki et al., 2017b; Hansen et al., 1997] to a policy iteration scheme. Our comparison on 31 continuous control tasks from parkour suite [Heess et al., 2017], DeepMind control suite [Tassa et al., 2018] and OpenAI Gym [Brockman et al., 2016] with diverse properties, limited amount of compute and a single set of hyperparameters, demonstrate the effectiveness of our method and the state of art results. Videos, summarizing results, can be found at goo.gl/HtvJKR .
A plethora of problems in AI, engineering and the sciences are naturally formalized as inference in discrete probabilistic models. Exact inference is often prohibitively expensive, as it may require evaluating the (unnormalized) target density on its entire domain. Here we consider the setting where only a limited budget of calls to the unnormalized density oracle is available, raising the challenge of where in the domain to allocate these function calls in order to construct a good approximate solution. We formulate this problem as an instance of sequential decision-making under uncertainty and leverage methods from reinforcement learning for probabilistic inference with budget constraints. In particular, we propose the TreeSample algorithm, an adaptation of Monte Carlo Tree Search to approximate inference. This algorithm caches all previous queries to the density oracle in an explicit search tree, and dynamically allocates new queries based on a "best-first" heuristic for exploration, using existing upper confidence bound methods. Our non-parametric inference method can be effectively combined with neural networks that compile approximate conditionals of the target, which are then used to guide the inference search and enable generalization across multiple target distributions. We show empirically that TreeSample outperforms standard approximate inference methods on synthetic factor graphs.
We propose to learn a kernel-based message operator which takes as input all expectation propagation (EP) incoming messages to a factor node and produces an outgoing message. In ordinary EP, computing an outgoing message involves estimating a multivariate integral which may not have an analytic expression. Learning such an operator allows one to bypass the expensive computation of the integral during inference by directly mapping all incoming messages into an outgoing message. The operator can be learned from training data (examples of input and output messages) which allows automated inference to be made on any kind of factor that can be sampled.
We propose an extension of the Restricted Boltzmann Machine (RBM) that allows the joint shape and appearance of foreground objects in cluttered images to be modeled independently of the background. We present a learning scheme that learns this representation directly from cluttered images with only very weak supervision. The model generates plausible samples and performs foreground-background segmentation. We demonstrate that representing foreground objects independently of the background can be beneficial in recognition tasks.
Stochastic computation graphs (SCGs) provide a formalism to represent structured optimization problems arising in artificial intelligence, including supervised, unsupervised, and reinforcement learning. Previous work has shown that an unbiased estimator of the gradient of the expected loss of SCGs can be derived from a single principle. However, this estimator often has high variance and requires a full model evaluation per data point, making this algorithm costly in large graphs. In this work, we address these problems by generalizing concepts from the reinforcement learning literature. We introduce the concepts of value functions, baselines and critics for arbitrary SCGs, and show how to use them to derive lower-variance gradient estimates from partial model evaluations, paving the way towards general and efficient credit assignment for gradient-based optimization. In doing so, we demonstrate how our results unify recent advances in the probabilistic inference and reinforcement learning literature.
We propose ThalNet, a deep learning model inspired by neocortical communication via the thalamus. Our model consists of recurrent neural modules that send features through a routing center, endowing the modules with the flexibility to share features over multiple time steps. We show that our model learns to route information hierarchically, processing input data by a chain of modules. We observe common architectures, such as feed forward neural networks and skip connections, emerging as special cases of our architecture, while novel connectivity patterns are learned for the text8 compression task. Our model outperforms standard recurrent neural networks on several sequential benchmarks.
In a variety of problems originating in supervised, unsupervised, and reinforcement learning, the loss function is defined by an expectation over a collection of random variables, which might be part of a probabilistic model or the external world. Estimating the gradient of this loss function, using samples, lies at the core of gradient-based learning algorithms for these problems. We introduce the formalism of stochastic computation graphs---directed acyclic graphs that include both deterministic functions and conditional probability distributions---and describe how to easily and automatically derive an unbiased estimator of the loss function's gradient. The resulting algorithm for computing the gradient estimator is a simple modification of the standard backpropagation algorithm. The generic scheme we propose unifies estimators derived in variety of prior work, along with variance-reduction techniques therein. It could assist researchers in developing intricate models involving a combination of stochastic and deterministic operations, enabling, for example, attention, memory, and control actions.
Applying convolutional neural networks to large images is computationally expensive because the amount of computation scales linearly with the number of image pixels. We present a novel recurrent neural network model that is capable of extracting information from an image or video by adaptively selecting a sequence of regions or locations and only processing the selected regions at high resolution. Like convolutional neural networks, the proposed model has a degree of translation invariance built-in, but the amount of computation it performs can be controlled independently of the input image size. While the model is non-differentiable, it can be trained using reinforcement learning methods to learn task-specific policies. We evaluate our model on several image classification tasks, where it significantly outperforms a convolutional neural network baseline on cluttered images, and on a dynamic visual control problem, where it learns to track a simple object without an explicit training signal for doing so.
Deep reinforcement learning (RL) algorithms have made great strides in recent years. An important remaining challenge is the ability to quickly transfer existing skills to novel tasks, and to combine existing skills with newly acquired ones. In domains where tasks are solved by composing skills this capacity holds the promise of dramatically reducing the data requirements of deep RL algorithms, and hence increasing their applicability. Recent work has studied ways of composing behaviors represented in the form of action-value functions. We analyze these methods to highlight their strengths and weaknesses, and point out situations where each of them is susceptible to poor performance. To perform this analysis we extend generalized policy improvement to the max-entropy framework and introduce a method for the practical implementation of successor features in continuous action spaces. Then we propose a novel approach which, in principle, recovers the optimal policy during transfer. This method works by explicitly learning the (discounted, future) divergence between policies. We study this approach in the tabular case and propose a scalable variant that is applicable in multi-dimensional continuous action spaces. We compare our approach with existing ones on a range of non-trivial continuous control problems with compositional structure, and demonstrate qualitatively better performance despite not requiring simultaneous observation of all task rewards.
Partially observed control problems are a challenging aspect of reinforcement learning. We extend two related, model-free algorithms for continuous control -- deterministic policy gradient and stochastic value gradient -- to solve partially observed domains using recurrent neural networks trained with backpropagation through time. We demonstrate that this approach, coupled with long-short term memory is able to solve a variety of physical control problems exhibiting an assortment of memory requirements. These include the short-term integration of information from noisy sensors and the identification of system parameters, as well as long-term memory problems that require preserving information over many time steps. We also demonstrate success on a combined exploration and memory problem in the form of a simplified version of the well-known Morris water maze task. Finally, we show that our approach can deal with high-dimensional observations by learning directly from pixels. We find that recurrent deterministic and stochastic policies are able to learn similarly good solutions to these tasks, including the water maze where the agent must learn effective search strategies.
We present an algorithm for learning an approximate action-value soft Q-function in the relative entropy regularised reinforcement learning setting, for which an optimal improved policy can be recovered in closed form. We use recent advances in normalising flows for parametrising the policy together with a learned value-function; and show how this combination can be used to implicitly represent Q-values of an arbitrary policy in continuous action space. Using simple temporal difference learning on the Q-values then leads to a unified objective for policy and value learning. We show how this approach considerably simplifies standard Actor-Critic off-policy algorithms, removing the need for a policy optimisation step. We perform experiments on a range of established reinforcement learning benchmarks, demonstrating that our approach allows for complex, multimodal policy distributions in continuous action spaces, while keeping the process of sampling from the policy both fast and exact.
Direct optimization is an appealing approach to differentiating through discrete quantities. Rather than relying on REINFORCE or continuous relaxations of discrete structures, it uses optimization in discrete space to compute gradients through a discrete argmax operation. In this paper, we develop reinforcement learning algorithms that use direct optimization to compute gradients of the expected return in environments with discrete actions. We call the resulting algorithms "direct policy gradient" algorithms and investigate their properties, showing that there is a built-in variance reduction technique and that a parameter that was previously viewed as a numerical approximation can be interpreted as controlling risk sensitivity. We also tackle challenges in algorithm design, leveraging ideas from A$^\star$ Sampling to develop a practical algorithm. Empirically, we show that the algorithm performs well in illustrative domains, and that it can make use of domain knowledge about upper bounds on return-to-go to speed up training.
Operating directly from raw high dimensional sensory inputs like images is still a challenge for robotic control. Recently, Reinforcement Learning methods have been proposed to solve specific tasks end-to-end, from pixels to torques. However, these approaches assume the access to a specified reward which may require specialized instrumentation of the environment. Furthermore, the obtained policy and representations tend to be task specific and may not transfer well. In this work we investigate completely self-supervised learning of a general image embedding and control primitives, based on finding the shortest time to reach any state. We also introduce a new structure for the state-action value function that builds a connection between model-free and model-based methods, and improves the performance of the learning algorithm. We experimentally demonstrate these findings in three simulated robotic tasks.
In this work, we consider the problem of autonomously discovering behavioral abstractions, or options, for reinforcement learning agents. We propose an algorithm that focuses on the termination condition, as opposed to -- as is common -- the policy. The termination condition is usually trained to optimize a control objective: an option ought to terminate if another has better value. We offer a different, information-theoretic perspective, and propose that terminations should focus instead on the compressibility of the option's encoding -- arguably a key reason for using abstractions. To achieve this algorithmically, we leverage the classical options framework, and learn the option transition model as a "critic" for the termination condition. Using this model, we derive gradients that optimize the desired criteria. We show that the resulting options are non-trivial, intuitively meaningful, and useful for learning and planning.
We study the emergence of cooperative behaviors in reinforcement learning agents by introducing a challenging competitive multi-agent soccer environment with continuous simulated physics. We demonstrate that decentralized, population-based training with co-play can lead to a progression in agents' behaviors: from random, to simple ball chasing, and finally showing evidence of cooperation. Our study highlights several of the challenges encountered in large scale multi-agent training in continuous control. In particular, we demonstrate that the automatic optimization of simple shaping rewards, not themselves conducive to co-operative behavior, can lead to long-horizon team behavior. We further apply an evaluation scheme, grounded by game theoretic principals, that can assess agent performance in the absence of pre-defined evaluation tasks or human baselines.
The naive application of Reinforcement Learning algorithms to continuous control problems -- such as locomotion and manipulation -- often results in policies which rely on high-amplitude, high-frequency control signals, known colloquially as bang-bang control. Although such solutions may indeed maximize task reward, they can be unsuitable for real world systems. Bang-bang control may lead to increased wear and tear or energy consumption, and tends to excite undesired second-order dynamics. To counteract this issue, multi-objective optimization can be used to simultaneously optimize both the reward and some auxiliary cost that discourages undesired (e.g. high-amplitude) control. In principle, such an approach can yield the sought after, smooth, control policies. It can, however, be hard to find the correct trade-off between cost and return that results in the desired behavior. In this paper we propose a new constraint-based reinforcement learning approach that ensures task success while minimizing one or more auxiliary costs (such as control effort). We employ Lagrangian relaxation to learn both (a) the parameters of a control policy that satisfies the desired constraints and (b) the Lagrangian multipliers for the optimization. Moreover, we demonstrate that we can satisfy constraints either in expectation or in a per-step fashion, and can even learn a single policy that is able to dynamically trade-off between return and cost. We demonstrate the efficacy of our approach using a number of continuous control benchmark tasks, a realistic, energy-optimized quadruped locomotion task, as well as a reaching task on a real robot arm.
Applying end-to-end learning to solve complex, interactive, pixel-driven control tasks on a robot is an unsolved problem. Deep Reinforcement Learning algorithms are too slow to achieve performance on a real robot, but their potential has been demonstrated in simulated environments. We propose using progressive networks to bridge the reality gap and transfer learned policies from simulation to the real world. The progressive net approach is a general framework that enables reuse of everything from low-level visual features to high-level policies for transfer to new tasks, enabling a compositional, yet simple, approach to building complex skills. We present an early demonstration of this approach with a number of experiments in the domain of robot manipulation that focus on bridging the reality gap. Unlike other proposed approaches, our real-world experiments demonstrate successful task learning from raw visual input on a fully actuated robot manipulator. Moreover, rather than relying on model-based trajectory optimisation, the task learning is accomplished using only deep reinforcement learning and sparse rewards.
We study a novel architecture and training procedure for locomotion tasks. A high-frequency, low-level "spinal" network with access to proprioceptive sensors learns sensorimotor primitives by training on simple tasks. This pre-trained module is fixed and connected to a low-frequency, high-level "cortical" network, with access to all sensors, which drives behavior by modulating the inputs to the spinal network. Where a monolithic end-to-end architecture fails completely, learning with a pre-trained spinal module succeeds at multiple high-level tasks, and enables the effective exploration required to learn from sparse rewards. We test our proposed architecture on three simulated bodies: a 16-dimensional swimming snake, a 20-dimensional quadruped, and a 54-dimensional humanoid. Our results are illustrated in the accompanying video at https://youtu.be/sboPYvhpraQ
We present a unified framework for learning continuous control policies using backpropagation. It supports stochastic control by treating stochasticity in the Bellman equation as a deterministic function of exogenous noise. The product is a spectrum of general policy gradient algorithms that range from model-free methods with value functions to model-based methods without value functions. We use learned models but only require observations from the environment in- stead of observations from model-predicted trajectories, minimizing the impact of compounded model errors. We apply these algorithms first to a toy stochastic control problem and then to several physics-based control problems in simulation. One of these variants, SVG(1), shows the effectiveness of learning models, value functions, and policies simultaneously in continuous domains.
Humans achieve efficient learning by relying on prior knowledge about the structure of naturally occurring tasks. There has been considerable interest in designing reinforcement learning algorithms with similar properties. This includes several proposals to learn the learning algorithm itself, an idea also referred to as meta learning. One formal interpretation of this idea is in terms of a partially observable multi-task reinforcement learning problem in which information about the task is hidden from the agent. Although agents that solve partially observable environments can be trained from rewards alone, shaping an agent's memory with additional supervision has been shown to boost learning efficiency. It is thus natural to ask what kind of supervision, if any, facilitates meta-learning. Here we explore several choices and develop an architecture that separates learning of the belief about the unknown task from learning of the policy, and that can be used effectively with privileged information about the task during training. We show that this approach can be very effective at solving standard meta-RL environments, as well as a complex continuous control environment in which a simulated robot has to execute various movement sequences.