scholarly journals Count-Based Exploration with the Successor Representation

2020 ◽  
Vol 34 (04) ◽  
pp. 5125-5133
Author(s):  
Marlos C. Machado ◽  
Marc G. Bellemare ◽  
Michael Bowling
Keyword(s):  
Reinforcement Learning ◽  
Function Approximation ◽  
State Of The Art ◽  
Transient Behavior ◽  
Simple Approach ◽  
Sample Complexity ◽  
Art Performance

In this paper we introduce a simple approach for exploration in reinforcement learning (RL) that allows us to develop theoretically justified algorithms in the tabular case but that is also extendable to settings where function approximation is required. Our approach is based on the successor representation (SR), which was originally introduced as a representation defining state generalization by the similarity of successor states. Here we show that the norm of the SR, while it is being learned, can be used as a reward bonus to incentivize exploration. In order to better understand this transient behavior of the norm of the SR we introduce the substochastic successor representation (SSR) and we show that it implicitly counts the number of times each state (or feature) has been observed. We use this result to introduce an algorithm that performs as well as some theoretically sample-efficient approaches. Finally, we extend these ideas to a deep RL algorithm and show that it achieves state-of-the-art performance in Atari 2600 games when in a low sample-complexity regime.

scholarly journals ERLP: Ensembles of Reinforcement Learning Policies (Student Abstract)

2020 ◽  
Vol 34 (10) ◽  
pp. 13905-13906
Author(s):  
Rohan Saphal ◽  
Balaraman Ravindran ◽  
Dheevatsa Mudigere ◽  
Sasikanth Avancha ◽  
Bharat Kaul
Keyword(s):  
Reinforcement Learning ◽  
State Of The Art ◽  
Multiple Models ◽  
Model Parameters ◽  
Continuous Control ◽  
Sample Complexity ◽  
Local Minima ◽  
Single Model ◽  
Learning Policies ◽  
Reinforcement Learning Models

Reinforcement learning algorithms are sensitive to hyper-parameters and require tuning and tweaking for specific environments for improving performance. Ensembles of reinforcement learning models on the other hand are known to be much more robust and stable. However, training multiple models independently on an environment suffers from high sample complexity. We present here a methodology to create multiple models from a single training instance that can be used in an ensemble through directed perturbation of the model parameters at regular intervals. This allows training a single model that converges to several local minima during the optimization process as a result of the perturbation. By saving the model parameters at each such instance, we obtain multiple policies during training that are ensembled during evaluation. We evaluate our approach on challenging discrete and continuous control tasks and also discuss various ensembling strategies. Our framework is substantially sample efficient, computationally inexpensive and is seen to outperform state of the art (SOTA) approaches

scholarly journals The Successful Ingredients of Policy Gradient Algorithms

2021 ◽  
Author(s):  
Sven Gronauer ◽  
Martin Gottwald ◽  
Klaus Diepold
Keyword(s):  
Reinforcement Learning ◽  
State Of The Art ◽  
Gradient Methods ◽  
Gradient Algorithms ◽  
The Sublime ◽  
Policy Gradient ◽  
Art Performance ◽  
Underlying Mechanisms

Despite the sublime success in recent years, the underlying mechanisms powering the advances of reinforcement learning are yet poorly understood. In this paper, we identify these mechanisms - which we call ingredients - in on-policy policy gradient methods and empirically determine their impact on the learning. To allow an equitable assessment, we conduct our experiments based on a unified and modular implementation. Our results underline the significance of recent algorithmic advances and demonstrate that reaching state-of-the-art performance may not need sophisticated algorithms but can also be accomplished by the combination of a few simple ingredients.

scholarly journals Multi-Task Deep Reinforcement Learning with PopArt

2019 ◽  
Vol 33 ◽  
pp. 3796-3803 ◽  
Author(s):  
Matteo Hessel ◽  
Hubert Soyer ◽  
Lasse Espeholt ◽  
Wojciech Czarnecki ◽  
Simon Schmitt ◽  
...  
Keyword(s):  
Reinforcement Learning ◽  
Human Performance ◽  
Learning Algorithm ◽  
State Of The Art ◽  
Single Agent ◽  
Learning System ◽  
Learning Platform ◽  
Art Performance ◽  
The One ◽  
First Time

The reinforcement learning (RL) community has made great strides in designing algorithms capable of exceeding human performance on specific tasks. These algorithms are mostly trained one task at the time, each new task requiring to train a brand new agent instance. This means the learning algorithm is general, but each solution is not; each agent can only solve the one task it was trained on. In this work, we study the problem of learning to master not one but multiple sequentialdecision tasks at once. A general issue in multi-task learning is that a balance must be found between the needs of multiple tasks competing for the limited resources of a single learning system. Many learning algorithms can get distracted by certain tasks in the set of tasks to solve. Such tasks appear more salient to the learning process, for instance because of the density or magnitude of the in-task rewards. This causes the algorithm to focus on those salient tasks at the expense of generality. We propose to automatically adapt the contribution of each task to the agent’s updates, so that all tasks have a similar impact on the learning dynamics. This resulted in state of the art performance on learning to play all games in a set of 57 diverse Atari games. Excitingly, our method learned a single trained policy - with a single set of weights - that exceeds median human performance. To our knowledge, this was the first time a single agent surpassed human-level performance on this multi-task domain. The same approach also demonstrated state of the art performance on a set of 30 tasks in the 3D reinforcement learning platform DeepMind Lab.

scholarly journals Reinforcement Learning with Perturbed Rewards

2020 ◽  
Vol 34 (04) ◽  
pp. 6202-6209 ◽  
Author(s):  
Jingkang Wang ◽  
Yang Liu ◽  
Bo Li
Keyword(s):  
Reinforcement Learning ◽  
Gaussian Noise ◽  
State Of The Art ◽  
Confusion Matrix ◽  
Error Rates ◽  
Average Score ◽  
Sample Complexity ◽  
Noisy Environments ◽  
The Core ◽  
Made In

Recent studies have shown that reinforcement learning (RL) models are vulnerable in various noisy scenarios. For instance, the observed reward channel is often subject to noise in practice (e.g., when rewards are collected through sensors), and is therefore not credible. In addition, for applications such as robotics, a deep reinforcement learning (DRL) algorithm can be manipulated to produce arbitrary errors by receiving corrupted rewards. In this paper, we consider noisy RL problems with perturbed rewards, which can be approximated with a confusion matrix. We develop a robust RL framework that enables agents to learn in noisy environments where only perturbed rewards are observed. Our solution framework builds on existing RL/DRL algorithms and firstly addresses the biased noisy reward setting without any assumptions on the true distribution (e.g., zero-mean Gaussian noise as made in previous works). The core ideas of our solution include estimating a reward confusion matrix and defining a set of unbiased surrogate rewards. We prove the convergence and sample complexity of our approach. Extensive experiments on different DRL platforms show that trained policies based on our estimated surrogate reward can achieve higher expected rewards, and converge faster than existing baselines. For instance, the state-of-the-art PPO algorithm is able to obtain 84.6% and 80.8% improvements on average score for five Atari games, with error rates as 10% and 30% respectively.

scholarly journals Don’t Do What Doesn’t Matter: Intrinsic Motivation with Action Usefulness

2021 ◽  
Author(s):  
Mathieu Seurin ◽  
Florian Strub ◽  
Philippe Preux ◽  
Olivier Pietquin
Keyword(s):  
Reinforcement Learning ◽  
Intrinsic Motivation ◽  
State Space ◽  
State Of The Art ◽  
The State ◽  
Sample Complexity ◽  
Art Methods ◽  
New States

Sparse rewards are double-edged training signals in reinforcement learning: easy to design but hard to optimize. Intrinsic motivation guidances have thus been developed toward alleviating the resulting exploration problem. They usually incentivize agents to look for new states through novelty signals. Yet, such methods encourage exhaustive exploration of the state space rather than focusing on the environment's salient interaction opportunities. We propose a new exploration method, called Don't Do What Doesn't Matter (DoWhaM), shifting the emphasis from state novelty to state with relevant actions. While most actions consistently change the state when used, e.g. moving the agent, some actions are only effective in specific states, e.g., opening a door, grabbing an object. DoWhaM detects and rewards actions that seldom affect the environment. We evaluate DoWhaM on the procedurally-generated environment MiniGrid against state-of-the-art methods. Experiments consistently show that DoWhaM greatly reduces sample complexity, installing the new state-of-the-art in MiniGrid.

scholarly journals Synthetic Experiences for Accelerating DQN Performance in Discrete Non-Deterministic Environments

Algorithms ◽  
2021 ◽  
Vol 14 (8) ◽  
pp. 226
Author(s):  
Wenzel Pilar von Pilchau ◽  
Anthony Stein ◽  
Jörg Hähner
Keyword(s):  
Reinforcement Learning ◽  
State Of The Art ◽  
Learning Algorithms ◽  
Weighted Average ◽  
Up States ◽  
Experience Replay

State-of-the-art Deep Reinforcement Learning Algorithms such as DQN and DDPG use the concept of a replay buffer called Experience Replay. The default usage contains only the experiences that have been gathered over the runtime. We propose a method called Interpolated Experience Replay that uses stored (real) transitions to create synthetic ones to assist the learner. In this first approach to this field, we limit ourselves to discrete and non-deterministic environments and use a simple equally weighted average of the reward in combination with observed follow-up states. We could demonstrate a significantly improved overall mean average in comparison to a DQN network with vanilla Experience Replay on the discrete and non-deterministic FrozenLake8x8-v0 environment.

scholarly journals Drone Deep Reinforcement Learning: A Review

Electronics ◽  
2021 ◽  
Vol 10 (9) ◽  
pp. 999
Author(s):  
Ahmad Taher Azar ◽  
Anis Koubaa ◽  
Nada Ali Mohamed ◽  
Habiba A. Ibrahim ◽  
Zahra Fathy Ibrahim ◽  
...  
Keyword(s):  
Reinforcement Learning ◽  
State Of The Art ◽  
Real Life ◽  
Environment Monitoring ◽  
Simulated Environments ◽  
Infrastructure Inspection ◽  
Remote Sensing Mapping ◽  
And Control ◽  
The Military ◽  
Uav Navigation

Unmanned Aerial Vehicles (UAVs) are increasingly being used in many challenging and diversified applications. These applications belong to the civilian and the military fields. To name a few; infrastructure inspection, traffic patrolling, remote sensing, mapping, surveillance, rescuing humans and animals, environment monitoring, and Intelligence, Surveillance, Target Acquisition, and Reconnaissance (ISTAR) operations. However, the use of UAVs in these applications needs a substantial level of autonomy. In other words, UAVs should have the ability to accomplish planned missions in unexpected situations without requiring human intervention. To ensure this level of autonomy, many artificial intelligence algorithms were designed. These algorithms targeted the guidance, navigation, and control (GNC) of UAVs. In this paper, we described the state of the art of one subset of these algorithms: the deep reinforcement learning (DRL) techniques. We made a detailed description of them, and we deduced the current limitations in this area. We noted that most of these DRL methods were designed to ensure stable and smooth UAV navigation by training computer-simulated environments. We realized that further research efforts are needed to address the challenges that restrain their deployment in real-life scenarios.

scholarly journals PyConvU-Net: a lightweight and multiscale network for biomedical image segmentation

2021 ◽  
Vol 22 (1) ◽  
Author(s):  
Changyong Li ◽  
Yongxian Fan ◽  
Xiaodong Cai
Keyword(s):  
Image Segmentation ◽  
Deep Learning ◽  
State Of The Art ◽  
Experimental Results ◽  
Actual Situation ◽  
Controlled Experiments ◽  
Biomedical Image ◽  
Segmentation Methods ◽  
Art Performance

Abstract Background With the development of deep learning (DL), more and more methods based on deep learning are proposed and achieve state-of-the-art performance in biomedical image segmentation. However, these methods are usually complex and require the support of powerful computing resources. According to the actual situation, it is impractical that we use huge computing resources in clinical situations. Thus, it is significant to develop accurate DL based biomedical image segmentation methods which depend on resources-constraint computing. Results A lightweight and multiscale network called PyConvU-Net is proposed to potentially work with low-resources computing. Through strictly controlled experiments, PyConvU-Net predictions have a good performance on three biomedical image segmentation tasks with the fewest parameters. Conclusions Our experimental results preliminarily demonstrate the potential of proposed PyConvU-Net in biomedical image segmentation with resources-constraint computing.

scholarly journals Contour Detection for Fibre of Preserved Szechuan Pickle Based on Dilated Convolution

2019 ◽  
Vol 9 (13) ◽  
pp. 2684 ◽  
Author(s):  
Hongyang Li ◽  
Lizhuang Liu ◽  
Zhenqi Han ◽  
Dan Zhao
Keyword(s):  
Edge Detection ◽  
State Of The Art ◽  
Contour Detection ◽  
Experimental Results ◽  
Contour Method ◽  
Class Differences ◽  
Dilated Convolution ◽  
The Mean ◽  
Art Performance

Peeling fibre is an indispensable process in the production of preserved Szechuan pickle, the accuracy of which can significantly influence the quality of the products, and thus the contour method of fibre detection, as a core algorithm of the automatic peeling device, is studied. The fibre contour is a kind of non-salient contour, characterized by big intra-class differences and small inter-class differences, meaning that the feature of the contour is not discriminative. The method called dilated-holistically-nested edge detection (Dilated-HED) is proposed to detect the fibre contour, which is built based on the HED network and dilated convolution. The experimental results for our dataset show that the Pixel Accuracy (PA) is 99.52% and the Mean Intersection over Union (MIoU) is 49.99%, achieving state-of-the-art performance.

Pruning by leveraging training dynamics

2021 ◽  
pp. 1-21
Author(s):  
Andrei C. Apostol ◽  
Maarten C. Stol ◽  
Patrick Forré
Keyword(s):  
State Of The Art ◽  
Object Classification ◽  
Compression Technique ◽  
Pruning Method ◽  
Model Compression ◽  
Art Performance

We propose a novel pruning method which uses the oscillations around 0, i.e. sign flips, that a weight has undergone during training in order to determine its saliency. Our method can perform pruning before the network has converged, requires little tuning effort due to having good default values for its hyperparameters, and can directly target the level of sparsity desired by the user. Our experiments, performed on a variety of object classification architectures, show that it is competitive with existing methods and achieves state-of-the-art performance for levels of sparsity of 99.6 % and above for 2 out of 3 of the architectures tested. Moreover, we demonstrate that our method is compatible with quantization, another model compression technique. For reproducibility, we release our code at https://github.com/AndreiXYZ/flipout.

Sign in / Sign up

Export Citation Format

Share Document