Searcharxiv⌕ Search

arXiv subjects

Olivier Sigaud

Publications and source records attributed to Olivier Sigaud.

70 records · Page 4Linked to original sources

The problem with DDPG: understanding failures in deterministic environments with sparse rewards

In environments with continuous state and action spaces, state-of-the-art actor-critic reinforcement learning algorithms can solve very complex problems, yet can also fail in environments that seem trivial, but the reason for such failures is still poorly understood. In this paper, we contribute a formal explanation of these failures in the particular case of sparse reward and deterministic environments. First, using a very elementary control problem, we illustrate that the learning process can get stuck into a fixed point corresponding to a poor solution. Then, generalizing from the studied example, we provide a detailed analysis of the underlying mechanisms which results in a new understanding of one of the convergence regimes of these algorithms. The resulting perspective casts a new light on already existing solutions to the issues we have highlighted, and suggests other potential approaches.

cs.LG↗

Policy Search in Continuous Action Domains: an Overview

Continuous action policy search is currently the focus of intensive research, driven both by the recent success of deep reinforcement learning algorithms and the emergence of competitors based on evolutionary algorithms. In this paper, we present a broad survey of policy search methods, providing a unified perspective on very different approaches, including also Bayesian Optimization and directed exploration methods. The main message of this overview is in the relationship between the families of methods, but we also outline some factors underlying sample efficiency properties of the various approaches.

cs.LG↗

CLIC: Curriculum Learning and Imitation for object Control in non-rewarding environments

In this paper we study a new reinforcement learning setting where the environment is non-rewarding, contains several possibly related objects of various controllability, and where an apt agent Bob acts independently, with non-observable intentions. We argue that this setting defines a realistic scenario and we present a generic discrete-state discrete-action model of such environments. To learn in this environment, we propose an unsupervised reinforcement learning agent called CLIC for Curriculum Learning and Imitation for Control. CLIC learns to control individual objects in its environment, and imitates Bob's interactions with these objects. It selects objects to focus on when training and imitating by maximizing its learning progress. We show that CLIC is an effective baseline in our new setting. It can effectively observe Bob to gain control of objects faster, even if Bob is not explicitly teaching. It can also follow Bob when he acts as a mentor and provides ordered demonstrations. Finally, when Bob controls objects that the agent cannot, or in presence of a hierarchy between objects in the environment, we show that CLIC ignores non-reproducible and already mastered interactions with objects, resulting in a greater benefit from imitation.

cs.LG↗

Investigating Generalisation in Continuous Deep Reinforcement Learning

Deep Reinforcement Learning has shown great success in a variety of control tasks. However, it is unclear how close we are to the vision of putting Deep RL into practice to solve real world problems. In particular, common practice in the field is to train policies on largely deterministic simulators and to evaluate algorithms through training performance alone, without a train/test distinction to ensure models generalise and are not overfitted. Moreover, it is not standard practice to check for generalisation under domain shift, although robustness to such system change between training and testing would be necessary for real-world Deep RL control, for example, in robotics. In this paper we study these issues by first characterising the sources of uncertainty that provide generalisation challenges in Deep RL. We then provide a new benchmark and thorough empirical evaluation of generalisation challenges for state of the art Deep RL methods. In particular, we show that, if generalisation is the goal, then common practice of evaluating algorithms based on their training performance leads to the wrong conclusions about algorithm choice. Finally, we evaluate several techniques for improving generalisation and draw conclusions about the most robust techniques to date.

cs.LG↗

CEM-RL: Combining evolutionary and gradient-based methods for policy search

Deep neuroevolution and deep reinforcement learning (deep RL) algorithms are two popular approaches to policy search. The former is widely applicable and rather stable, but suffers from low sample efficiency. By contrast, the latter is more sample efficient, but the most sample efficient variants are also rather unstable and highly sensitive to hyper-parameter setting. So far, these families of methods have mostly been compared as competing tools. However, an emerging approach consists in combining them so as to get the best of both worlds. Two previously existing combinations use either an ad hoc evolutionary algorithm or a goal exploration process together with the Deep Deterministic Policy Gradient (DDPG) algorithm, a sample efficient off-policy deep RL algorithm. In this paper, we propose a different combination scheme using the simple cross-entropy method (CEM) and Twin Delayed Deep Deterministic policy gradient (td3), another off-policy deep RL algorithm which improves over ddpg. We evaluate the resulting method, cem-rl, on a set of benchmarks classically used in deep RL. We show that cem-rl benefits from several advantages over its competitors and offers a satisfactory trade-off between performance and sample efficiency.

cs.LG↗

Identification of Invariant Sensorimotor Structures as a Prerequisite for the Discovery of Objects

Perceiving the surrounding environment in terms of objects is useful for any general purpose intelligent agent. In this paper, we investigate a fundamental mechanism making object perception possible, namely the identification of spatio-temporally invariant structures in the sensorimotor experience of an agent. We take inspiration from the Sensorimotor Contingencies Theory to define a computational model of this mechanism through a sensorimotor, unsupervised and predictive approach. Our model is based on processing the unsupervised interaction of an artificial agent with its environment. We show how spatio-temporally invariant structures in the environment induce regularities in the sensorimotor experience of an agent, and how this agent, while building a predictive model of its sensorimotor experience, can capture them as densely connected subgraphs in a graph of sensory states connected by motor commands. Our approach is focused on elementary mechanisms, and is illustrated with a set of simple experiments in which an agent interacts with an environment. We show how the agent can build an internal model of moving but spatio-temporally invariant structures by performing a Spectral Clustering of the graph modeling its overall sensorimotor experiences. We systematically examine properties of the model, shedding light more globally on the specificities of the paradigm with respect to methods based on the supervised processing of collections of static images.

cs.AI↗

Unsupervised Learning of Goal Spaces for Intrinsically Motivated Goal Exploration

Intrinsically motivated goal exploration algorithms enable machines to discover repertoires of policies that produce a diversity of effects in complex environments. These exploration algorithms have been shown to allow real world robots to acquire skills such as tool use in high-dimensional continuous state and action spaces. However, they have so far assumed that self-generated goals are sampled in a specifically engineered feature space, limiting their autonomy. In this work, we propose to use deep representation learning algorithms to learn an adequate goal space. This is a developmental 2-stage approach: first, in a perceptual learning stage, deep learning algorithms use passive raw sensor observations of world changes to learn a corresponding latent space; then goal exploration happens in a second stage by sampling goals in this latent space. We present experiments where a simulated robot arm interacts with an object, and we show that exploration algorithms using such learned representations can match the performance obtained using engineered representations.

cs.LG↗

Accuracy-based Curriculum Learning in Deep Reinforcement Learning

In this paper, we investigate a new form of automated curriculum learning based on adaptive selection of accuracy requirements, called accuracy-based curriculum learning. Using a reinforcement learning agent based on the Deep Deterministic Policy Gradient algorithm and addressing the Reacher environment, we first show that an agent trained with various accuracy requirements sampled randomly learns more efficiently than when asked to be very accurate at all times. Then we show that adaptive selection of accuracy requirements, based on a local measure of competence progress, automatically generates a curriculum where difficulty progressively increases, resulting in a better learning efficiency than sampling randomly.

cs.LG↗

GEP-PG: Decoupling Exploration and Exploitation in Deep Reinforcement Learning Algorithms

In continuous action domains, standard deep reinforcement learning algorithms like DDPG suffer from inefficient exploration when facing sparse or deceptive reward problems. Conversely, evolutionary and developmental methods focusing on exploration like Novelty Search, Quality-Diversity or Goal Exploration Processes explore more robustly but are less efficient at fine-tuning policies using gradient descent. In this paper, we present the GEP-PG approach, taking the best of both worlds by sequentially combining a Goal Exploration Process and two variants of DDPG. We study the learning performance of these components and their combination on a low dimensional deceptive reward problem and on the larger Half-Cheetah benchmark. We show that DDPG fails on the former and that GEP-PG improves over the best DDPG variant in both environments. Supplementary videos and discussion can be found at http://frama.link/gep_pg, the code at http://github.com/flowersteam/geppg.

cs.LG↗

Importance mixing: Improving sample reuse in evolutionary policy search methods

Deep neuroevolution, that is evolutionary policy search methods based on deep neural networks, have recently emerged as a competitor to deep reinforcement learning algorithms due to their better parallelization capabilities. However, these methods still suffer from a far worse sample efficiency. In this paper we investigate whether a mechanism known as "importance mixing" can significantly improve their sample efficiency. We provide a didactic presentation of importance mixing and we explain how it can be extended to reuse more samples. Then, from an empirical comparison based on a simple benchmark, we show that, though it actually provides better sample efficiency, it is still far from the sample efficiency of deep reinforcement learning, though it is more stable.

cs.LG↗

How Many Random Seeds? Statistical Power Analysis in Deep Reinforcement Learning Experiments

Consistently checking the statistical significance of experimental results is one of the mandatory methodological steps to address the so-called "reproducibility crisis" in deep reinforcement learning. In this tutorial paper, we explain how the number of random seeds relates to the probabilities of statistical errors. For both the t-test and the bootstrap confidence interval test, we recall theoretical guidelines to determine the number of random seeds one should use to provide a statistically significant comparison of the performance of two algorithms. Finally, we discuss the influence of deviations from the assumptions usually made by statistical tests. We show that they can lead to inaccurate evaluations of statistical errors and provide guidelines to counter these negative effects. We make our code available to perform the tests.

cs.LG↗

Actor-critic versus direct policy search: a comparison based on sample complexity

Sample efficiency is a critical property when optimizing policy parameters for the controller of a robot. In this paper, we evaluate two state-of-the-art policy optimization algorithms. One is a recent deep reinforcement learning method based on an actor-critic algorithm, Deep Deterministic Policy Gradient (DDPG), that has been shown to perform well on various control benchmarks. The other one is a direct policy search method, Covariance Matrix Adaptation Evolution Strategy (CMA-ES), a black-box optimization method that is widely used for robot learning. The algorithms are evaluated on a continuous version of the mountain car benchmark problem, so as to compare their sample complexity. From a preliminary analysis, we expect DDPG to be more sample efficient than CMA-ES, which is confirmed by our experimental results.

cs.LG↗

Gated networks: an inventory

Gated networks are networks that contain gating connections, in which the outputs of at least two neurons are multiplied. Initially, gated networks were used to learn relationships between two input sources, such as pixels from two images. More recently, they have been applied to learning activity recognition or multi-modal representations. The aims of this paper are threefold: 1) to explain the basic computations in gated networks to the non-expert, while adopting a standpoint that insists on their symmetric nature. 2) to serve as a quick reference guide to the recent literature, by providing an inventory of applications of these networks, as well as recent extensions to the basic architecture. 3) to suggest future research directions and applications.

cs.LG↗

Chi-square Tests Driven Method for Learning the Structure of Factored MDPs

SDYNA is a general framework designed to address large stochastic reinforcement learning problems. Unlike previous model based methods in FMDPs, it incrementally learns the structure and the parameters of a RL problem using supervised learning techniques. Then, it integrates decision-theoric planning algorithms based on FMDPs to compute its policy. SPITI is an instanciation of SDYNA that exploits ITI, an incremental decision tree algorithm, to learn the reward function and the Dynamic Bayesian Networks with local structures representing the transition function of the problem. These representations are used by an incremental version of the Structured Value Iteration algorithm. In order to learn the structure, SPITI uses Chi-Square tests to detect the independence between two probability distributions. Thus, we study the relation between the threshold used in the Chi-Square test, the size of the model built and the relative error of the value function of the induced policy with respect to the optimal value. We show that, on stochastic problems, one can tune the threshold so as to generate both a compact model and an efficient policy. Then, we show that SPITI, while keeping its model compact, uses the generalization property of its learning method to perform better than a stochastic classical tabular algorithm in large RL problem with an unknown structure. We also introduce a new measure based on Chi-Square to qualify the accuracy of the model learned by SPITI. We qualitatively show that the generalization property in SPITI within the FMDP framework may prevent an exponential growth of the time required to learn the structure of large stochastic RL problems.

cs.LG↗

Path Integral Policy Improvement with Covariance Matrix Adaptation

There has been a recent focus in reinforcement learning on addressing continuous state and action problems by optimizing parameterized policies. PI2 is a recent example of this approach. It combines a derivation from first principles of stochastic optimal control with tools from statistical estimation theory. In this paper, we consider PI2 as a member of the wider family of methods which share the concept of probability-weighted averaging to iteratively update parameters to optimize a cost function. We compare PI2 to other members of the same family - Cross-Entropy Methods and CMAES - at the conceptual level and in terms of performance. The comparison suggests the derivation of a novel algorithm which we call PI2-CMA for "Path Integral Policy Improvement with Covariance Matrix Adaptation". PI2-CMA's main advantage is that it determines the magnitude of the exploration noise automatically.

cs.LG↗

ATNoSFERES revisited

ATNoSFERES is a Pittsburgh style Learning Classifier System (LCS) in which the rules are represented as edges of an Augmented Transition Network. Genotypes are strings of tokens of a stack-based language, whose execution builds the labeled graph. The original ATNoSFERES, using a bitstring to represent the language tokens, has been favorably compared in previous work to several Michigan style LCSs architectures in the context of Non Markov problems. Several modifications of ATNoSFERES are proposed here: the most important one conceptually being a representational change: each token is now represented by an integer, hence the genotype is a string of integers; several other modifications of the underlying grammar language are also proposed. The resulting ATNoSFERES-II is validated on several standard animat Non Markov problems, on which it outperforms all previously published results in the LCS literature. The reasons for these improvement are carefully analyzed, and some assumptions are proposed on the underlying mechanisms in order to explain these good results.

cs.AI↗