arXiv subjects

Matthieu Geist

Publications and source records attributed to Matthieu Geist.

At least 109 records · Page 6Linked to original sources

arXiv2020

Modified Actor-Critics

Recent successful deep reinforcement learning algorithms, such as Trust Region Policy Optimization (TRPO) or Proximal Policy Optimization (PPO), are fundamentally variations of conservative policy iteration (CPI). These algorithms iterate policy evaluation followed by a softened policy improvement step. As so, they are naturally on-policy. In this paper, we propose to combine (any kind of) soft greediness with Modified Policy Iteration (MPI). The proposed abstract framework applies repeatedly: (i) a partial policy evaluation step that allows off-policy learning and (ii) any softened greedy step. Our contribution can be seen as a new generic tool for the deep reinforcement learning toolbox. As a proof of concept, we instantiate this framework with the PPO greediness. Comparison to the original PPO shows that our algorithm is much more sample efficient. We also show that it is competitive with the state-of-art off-policy algorithm Soft Actor Critic (SAC).

cs.LG↗

arXiv2020

CopyCAT: Taking Control of Neural Policies with Constant Attacks

We propose a new perspective on adversarial attacks against deep reinforcement learning agents. Our main contribution is CopyCAT, a targeted attack able to consistently lure an agent into following an outsider's policy. It is pre-computed, therefore fast inferred, and could thus be usable in a real-time scenario. We show its effectiveness on Atari 2600 games in the novel read-only setting. In this setting, the adversary cannot directly modify the agent's state -- its representation of the environment -- but can only attack the agent's observation -- its perception of the environment. Directly modifying the agent's state would require a write-access to the agent's inner workings and we argue that this assumption is too strong in realistic settings.

cs.LG↗

arXiv2020

Deep Conservative Policy Iteration

Conservative Policy Iteration (CPI) is a founding algorithm of Approximate Dynamic Programming (ADP). Its core principle is to stabilize greediness through stochastic mixtures of consecutive policies. It comes with strong theoretical guarantees, and inspired approaches in deep Reinforcement Learning (RL). However, CPI itself has rarely been implemented, never with neural networks, and only experimented on toy problems. In this paper, we show how CPI can be practically combined with deep RL with discrete actions. We also introduce adaptive mixture rates inspired by the theory. We experiment thoroughly the resulting algorithm on the simple Cartpole problem, and validate the proposed method on a representative subset of Atari games. Overall, this work suggests that revisiting classic ADP may lead to improved and more stable deep RL algorithms.

cs.LG↗

arXiv2019

Self-Attentional Credit Assignment for Transfer in Reinforcement Learning

The ability to transfer knowledge to novel environments and tasks is a sensible desiderata for general learning agents. Despite the apparent promises, transfer in RL is still an open and little exploited research area. In this paper, we take a brand-new perspective about transfer: we suggest that the ability to assign credit unveils structural invariants in the tasks that can be transferred to make RL more sample-efficient. Our main contribution is SECRET, a novel approach to transfer learning for RL that uses a backward-view credit assignment mechanism based on a self-attentive architecture. Two aspects are key to its generality: it learns to assign credit as a separate offline supervised process and exclusively modifies the reward function. Consequently, it can be supplemented by transfer methods that do not modify the reward function and it can be plugged on top of any RL algorithm.

cs.LG↗

arXiv2019

On Connections between Constrained Optimization and Reinforcement Learning

Dynamic Programming (DP) provides standard algorithms to solve Markov Decision Processes. However, these algorithms generally do not optimize a scalar objective function. In this paper, we draw connections between DP and (constrained) convex optimization. Specifically, we show clear links in the algorithmic structure between three DP schemes and optimization algorithms. We link Conservative Policy Iteration to Frank-Wolfe, Mirror-Descent Modified Policy Iteration to Mirror Descent, and Politex (Policy Iteration Using Expert Prediction) to Dual Averaging. These abstract DP schemes are representative of a number of (deep) Reinforcement Learning (RL) algorithms. By highlighting these connections (most of which have been noticed earlier, but in a scattered way), we would like to encourage further studies linking RL and convex optimization, that could lead to the design of new, more efficient, and better understood RL algorithms.

cs.LG↗

arXiv2019

Semi-Supervised Domain Adaptation with Representation Learning for Semantic Segmentation across Time

Deep learning generates state-of-the-art semantic segmentation provided that a large number of images together with pixel-wise annotations are available. To alleviate the expensive data collection process, we propose a semi-supervised domain adaptation method for the specific case of images with similar semantic content but different pixel distributions. A network trained with supervision on a past dataset is finetuned on the new dataset to conserve its features maps. The domain adaptation becomes a simple regression between feature maps and does not require annotations on the new dataset. This method reaches performances similar to classic transfer learning on the PASCAL VOC dataset with synthetic transformations.

cs.CV↗

arXiv2019

Learning Sensor Placement from Demonstration for UAV networks

This work demonstrates how to leverage previous network expert demonstrations of UAV deployment to automate the drones placement in civil applications. Optimal UAV placement is an NP-complete problem: it requires a closed-form utility function that defines the environment and the UAV constraints, it is not unique and must be defined for each new UAV mission. This complex and time-consuming process hinders the development of UAV-networks in civil applications. We propose a method that leverages previous network expert solutions of UAV-network deployment to learn the expert's untold utility function form demonstrations only. This is especially interesting as it may be difficult for the inspection expert to explicit his expertise into such a function as it is too complex. Once learned, our model generates a utility function which maxima match expert UAV locations. We test this method on a Wi-Fi UAV network application inside a crowd simulator and reach similar quality-of-service as the expert. We show that our method is not limited to this UAV application and can be extended to other missions such as building monitoring.

stat.AP↗

arXiv2019

ELF: Embedded Localisation of Features in pre-trained CNN

This paper introduces a novel feature detector based only on information embedded inside a CNN trained on standard tasks (e.g. classification). While previous works already show that the features of a trained CNN are suitable descriptors, we show here how to extract the feature locations from the network to build a detector. This information is computed from the gradient of the feature map with respect to the input image. This provides a saliency map with local maxima on relevant keypoint locations. Contrary to recent CNN-based detectors, this method requires neither supervised training nor finetuning. We evaluate how repeatable and how matchable the detected keypoints are with the repeatability and matching scores. Matchability is measured with a simple descriptor introduced for the sake of the evaluation. This novel detector reaches similar performances on the standard evaluation HPatches dataset, as well as comparable robustness against illumination and viewpoint changes on Webcam and photo-tourism images. These results show that a CNN trained on a standard task embeds feature location information that is as relevant as when the CNN is specifically trained for feature detection.

cs.CV↗

arXiv2019

MULEX: Disentangling Exploitation from Exploration in Deep RL

An agent learning through interactions should balance its action selection process between probing the environment to discover new rewards and using the information acquired in the past to adopt useful behaviour. This trade-off is usually obtained by perturbing either the agent's actions (e.g., e-greedy or Gibbs sampling) or the agent's parameters (e.g., NoisyNet), or by modifying the reward it receives (e.g., exploration bonus, intrinsic motivation, or hand-shaped rewards). Here, we adopt a disruptive but simple and generic perspective, where we explicitly disentangle exploration and exploitation. Different losses are optimized in parallel, one of them coming from the true objective (maximizing cumulative rewards from the environment) and others being related to exploration. Every loss is used in turn to learn a policy that generates transitions, all shared in a single replay buffer. Off-policy methods are then applied to these transitions to optimize each loss. We showcase our approach on a hard-exploration environment, show its sample-efficiency and robustness, and discuss further implications.

cs.LG↗

arXiv2019

A Theory of Regularized Markov Decision Processes

Many recent successful (deep) reinforcement learning algorithms make use of regularization, generally based on entropy or Kullback-Leibler divergence. We propose a general theory of regularized Markov Decision Processes that generalizes these approaches in two directions: we consider a larger class of regularizers, and we consider the general modified policy iteration approach, encompassing both policy iteration and value iteration. The core building blocks of this theory are a notion of regularized Bellman operator and the Legendre-Fenchel transform, a classical tool of convex optimization. This approach allows for error propagation analyses of general algorithmic schemes of which (possibly variants of) classical algorithms such as Trust Region Policy Optimization, Soft Q-learning, Stochastic Actor Critic or Dynamic Policy Programming are special cases. This also draws connections to proximal convex optimization, especially to Mirror Descent.

cs.LG↗

arXiv2018

Image-based Natural Language Understanding Using 2D Convolutional Neural Networks

We propose a new approach to natural language understanding in which we consider the input text as an image and apply 2D Convolutional Neural Networks to learn the local and global semantics of the sentences from the variations ofthe visual patterns of words. Our approach demonstrates that it is possible to get semantically meaningful features from images with text without using optical character recognition and sequential processing pipelines, techniques that traditional Natural Language Understanding algorithms require. To validate our approach, we present results for two applications: text classification and dialog modeling. Using a 2D Convolutional Neural Network, we were able to outperform the state-of-art accuracy results of non-Latin alphabet-based text classification and achieved promising results for eight text classification datasets. Furthermore, our approach outperformed the memory networks when using out of vocabulary entities fromtask 4 of the bAbI dialog dataset.

cs.CL↗

arXiv2018

Anderson Acceleration for Reinforcement Learning

Anderson acceleration is an old and simple method for accelerating the computation of a fixed point. However, as far as we know and quite surprisingly, it has never been applied to dynamic programming or reinforcement learning. In this paper, we explain briefly what Anderson acceleration is and how it can be applied to value iteration, this being supported by preliminary experiments showing a significant speed up of convergence, that we critically discuss. We also discuss how this idea could be applied more generally to (deep) reinforcement learning.

cs.LG↗

arXiv2018

Human Activity Recognition using Recurrent Neural Networks

Human activity recognition using smart home sensors is one of the bases of ubiquitous computing in smart environments and a topic undergoing intense research in the field of ambient assisted living. The increasingly large amount of data sets calls for machine learning methods. In this paper, we introduce a deep learning model that learns to classify human activities without using any prior knowledge. For this purpose, a Long Short Term Memory (LSTM) Recurrent Neural Network was applied to three real world smart home datasets. The results of these experiments show that the proposed approach outperforms the existing ones in terms of accuracy and performance.

cs.NE↗

arXiv2018

A Deep Learning Approach for Privacy Preservation in Assisted Living

In the era of Internet of Things (IoT) technologies the potential for privacy invasion is becoming a major concern especially in regards to healthcare data and Ambient Assisted Living (AAL) environments. Systems that offer AAL technologies make extensive use of personal data in order to provide services that are context-aware and personalized. This makes privacy preservation a very important issue especially since the users are not always aware of the privacy risks they could face. A lot of progress has been made in the deep learning field, however, there has been lack of research on privacy preservation of sensitive personal data with the use of deep learning. In this paper we focus on a Long Short Term Memory (LSTM) Encoder-Decoder, which is a principal component of deep learning, and propose a new encoding technique that allows the creation of different AAL data views, depending on the access level of the end user and the information they require access to. The efficiency and effectiveness of the proposed method are demonstrated with experiments on a simulated AAL dataset. Qualitatively, we show that the proposed model learns privacy operations such as disclosure, deletion and generalization and can perform encoding and decoding of the data with almost perfect recovery.

eess.SP↗

arXiv2017

Is the Bellman residual a bad proxy?

This paper aims at theoretically and empirically comparing two standard optimization criteria for Reinforcement Learning: i) maximization of the mean value and ii) minimization of the Bellman residual. For that purpose, we place ourselves in the framework of policy search algorithms, that are usually designed to maximize the mean value, and derive a method that minimizes the residual $\|T_* v_π- v_π\|_{1,ν}$ over policies. A theoretical analysis shows how good this proxy is to policy optimization, and notably that it is better than its value-based counterpart. We also propose experiments on randomly generated generic Markov decision processes, specifically designed for studying the influence of the involved concentrability coefficient. They show that the Bellman residual is generally a bad proxy to policy optimization and that directly maximizing the mean value is much better, despite the current lack of deep theoretical analysis. This might seem obvious, as directly addressing the problem of interest is usually better, but given the prevalence of (projected) Bellman residual minimization in value-based reinforcement learning, we believe that this question is worth to be considered.

cs.LG↗

arXiv2016

Difference of Convex Functions Programming Applied to Control with Expert Data

This paper reports applications of Difference of Convex functions (DC) programming to Learning from Demonstrations (LfD) and Reinforcement Learning (RL) with expert data. This is made possible because the norm of the Optimal Bellman Residual (OBR), which is at the heart of many RL and LfD algorithms, is DC. Improvement in performance is demonstrated on two specific algorithms, namely Reward-regularized Classification for Apprenticeship Learning (RCAL) and Reinforcement Learning with Expert Demonstrations (RLED), through experiments on generic Markov Decision Processes (MDP), called Garnets.

math.OC↗

arXiv2014

Kalman Temporal Differences

Because reinforcement learning suffers from a lack of scalability, online value (and Q-) function approximation has received increasing interest this last decade. This contribution introduces a novel approximation scheme, namely the Kalman Temporal Differences (KTD) framework, that exhibits the following features: sample-efficiency, non-linear approximation, non-stationarity handling and uncertainty management. A first KTD-based algorithm is provided for deterministic Markov Decision Processes (MDP) which produces biased estimates in the case of stochastic transitions. Than the eXtended KTD framework (XKTD), solving stochastic MDP, is described. Convergence is analyzed for special cases for both deterministic and stochastic transitions. Related algorithms are experimented on classical benchmarks. They compare favorably to the state of the art while exhibiting the announced features.

cs.LG↗

arXiv2013

Policy Search: Any Local Optimum Enjoys a Global Performance Guarantee

Local Policy Search is a popular reinforcement learning approach for handling large state spaces. Formally, it searches locally in a paramet erized policy space in order to maximize the associated value function averaged over some predefined distribution. It is probably commonly b elieved that the best one can hope in general from such an approach is to get a local optimum of this criterion. In this article, we show th e following surprising result: \emph{any} (approximate) \emph{local optimum} enjoys a \emph{global performance guarantee}. We compare this g uarantee with the one that is satisfied by Direct Policy Iteration, an approximate dynamic programming algorithm that does some form of Poli cy Search: if the approximation error of Local Policy Search may generally be bigger (because local search requires to consider a space of s tochastic policies), we argue that the concentrability coefficient that appears in the performance bound is much nicer. Finally, we discuss several practical and theoretical consequences of our analysis.

cs.LG↗