Filters








25,057 Hits in 4.6 sec

On the Practical Consistency of Meta-Reinforcement Learning Algorithms [article]

Zheng Xiong, Luisa Zintgraf, Jacob Beck, Risto Vuorio, Shimon Whiteson
2021 arXiv   pre-print
Consistency is the theoretical property of a meta learning algorithm that ensures that, under certain assumptions, it can adapt to any task at test time.  ...  In this paper, we empirically investigate this question on a set of representative meta-RL algorithms.  ...  Acknowledgments and Disclosure of Funding  ... 
arXiv:2112.00478v1 fatcat:esoasqbwajddzkqhhum536ijeu

Guided Meta-Policy Search [article]

Russell Mendonca, Abhishek Gupta, Rosen Kralev, Pieter Abbeel, Sergey Levine, Chelsea Finn
2020 arXiv   pre-print
Reinforcement learning (RL) algorithms have demonstrated promising results on complex tasks, yet often require impractical numbers of samples since they learn from scratch.  ...  However, in practice, these algorithms generally also require large amounts of on-policy experience during the meta-training process, making them impractical for use in many problems.  ...  Acknowledgements The authors would like to thank Tianhe Yu for contributions on an early version of the paper.  ... 
arXiv:1904.00956v2 fatcat:wq6jlo3vfzeizhl75k7lxhfjnq

Meta-Reinforcement Learning Robust to Distributional Shift via Model Identification and Experience Relabeling [article]

Russell Mendonca, Xinyang Geng, Chelsea Finn, Sergey Levine
2020 arXiv   pre-print
Reinforcement learning algorithms can acquire policies for complex tasks autonomously. However, the number of samples required to learn a diverse set of skills can be prohibitively large.  ...  While meta-reinforcement learning methods have enabled agents to leverage prior experience to adapt quickly to new tasks, their performance depends crucially on how close the new task is to the previously  ...  Related Work Meta-reinforcement learning algorithms extend the framework of meta-learning [28, 34, 22, 1] to the reinforcement learning setting.  ... 
arXiv:2006.07178v2 fatcat:h3u4q2nuqjc2harv7tp4vcin24

Adaptive Adversarial Training for Meta Reinforcement Learning [article]

Shiqi Chen, Zhengyu Chen, Donglin Wang
2021 arXiv   pre-print
Meta Reinforcement Learning (MRL) enables an agent to learn from a limited number of past trajectories and extrapolate to a new task. In this paper, we attempt to improve the robustness of MRL.  ...  That allows us to enhance the robustness of MRL to adversal attacks by leveraging these attacks during meta training process.  ...  The authors gratefully acknowledge funding support from the Westlake University and Bright Dream Joint Institute for Intelligent Robotics.  ... 
arXiv:2104.13302v1 fatcat:qhutmcukwresfd2zki2zq4dswu

Meta-Gradient Reinforcement Learning [article]

Zhongwen Xu, Hado van Hasselt, David Silver
2018 arXiv   pre-print
The goal of reinforcement learning algorithms is to estimate and/or optimise the value function.  ...  Instead, the majority of reinforcement learning algorithms estimate and/or optimise a proxy for the value function.  ...  for their suggestions and comments on an early version of the paper.  ... 
arXiv:1805.09801v1 fatcat:mls5nqcgprbcpkdazc7fmnsuk4

Improving Generalization in Meta Reinforcement Learning using Learned Objectives [article]

Louis Kirsch, Sjoerd van Steenkiste, Jürgen Schmidhuber
2020 arXiv   pre-print
Biological evolution has distilled the experiences of many learners into the general learning algorithms of humans.  ...  Our novel meta reinforcement learning algorithm MetaGenRL is inspired by this process.  ...  We also thank NVIDIA Corporation for donating a DGX-1 as part of the Pioneers of AI Research Award and to IBM for donating a Minsky machine.  ... 
arXiv:1910.04098v2 fatcat:hdqyxrvuercu5hixjvak5v4pcq

Unsupervised Reinforcement Learning of Transferable Meta-Skills for Embodied Navigation

Juncheng Li, Xin Wang, Siliang Tang, Haizhou Shi, Fei Wu, Yueting Zhuang, William Yang Wang
2020 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)  
A key challenge for current deep reinforcement learning models lies in the requirements for a large amount of training data.  ...  We propose a novel unsupervised reinforcement learning approach to learn transferable meta-skills (e.g., bypass obstacles, go straight) from unannotated environments without any supervisory signals.  ...  Meta-Reinforcement Learning on the Proposed Tasks Inspired by meta-learning algorithms [28, 9, 14, 31, 11 ] that leverage experience across many tasks to learn new tasks quickly and efficiently, our method  ... 
doi:10.1109/cvpr42600.2020.01214 dblp:conf/cvpr/00060TS0ZW20 fatcat:esteyckgmvcxdkiziegilsiqqm

Unsupervised Reinforcement Learning of Transferable Meta-Skills for Embodied Navigation [article]

Juncheng Li, Xin Wang, Siliang Tang, Haizhou Shi, Fei Wu, Yueting Zhuang, William Yang Wang
2020 arXiv   pre-print
A key challenge for current deep reinforcement learning models lies in the requirements for a large amount of training data.  ...  We propose a novel unsupervised reinforcement learning approach to learn transferable meta-skills (e.g., bypass obstacles, go straight) from unannotated environments without any supervisory signals.  ...  Meta-Reinforcement Learning on the Proposed Tasks Inspired by meta-learning algorithms [28, 9, 14, 31, 11 ] that leverage experience across many tasks to learn new tasks quickly and efficiently, our method  ... 
arXiv:1911.07450v2 fatcat:pqja4fndzjai5f3zkmfzvxanyy

Meta-SGD: Learning to Learn Quickly for Few-Shot Learning [article]

Zhenguo Li, Fengwei Zhou, Fei Chen, Hang Li
2017 arXiv   pre-print
Meta-SGD shows highly competitive performance for few-shot learning on regression, classification, and reinforcement learning.  ...  In contrast, meta-learning learns from many related tasks a meta-learner that can learn a new task more accurately and faster with fewer examples, where the choice of meta-learners is crucial.  ...  Experimental Results We evaluate the proposed meta-learner Meta-SGD on a variety of few-shot learning problems on regression, classification, and reinforcement learning.  ... 
arXiv:1707.09835v2 fatcat:lizlpnejxvbgbflsaupj4p6euy

REIN-2: Giving Birth to Prepared Reinforcement Learning Agents Using Reinforcement Learning Agents [article]

Aristotelis Lazaridis, Ioannis Vlahavas
2021 arXiv   pre-print
Deep Reinforcement Learning (Deep RL) has been in the spotlight for the past few years, due to its remarkable abilities to solve problems which were considered to be practically unsolvable using traditional  ...  In an effort to patch these issues, we integrated a meta-learning technique in order to shift the objective of learning to solve a task into the objective of learning how to learn to solve a task (or a  ...  However, this vector has a relatively huge length, consisting of thousands of parameters, which makes the learning process for the meta-learner even more difficult.  ... 
arXiv:2110.05128v1 fatcat:2betqtk5anhznelqmexba6574y

Meta-Gradient Reinforcement Learning with an Objective Discovered Online [article]

Zhongwen Xu, Hado van Hasselt, Matteo Hessel, Junhyuk Oh, Satinder Singh, David Silver
2020 arXiv   pre-print
On the Atari Learning Environment, the meta-gradient algorithm adapts over time to learn with greater efficiency, eventually outperforming the median score of a strong actor-critic baseline.  ...  Deep reinforcement learning includes a broad family of algorithms that parameterise an internal representation, such as a value function or policy, by a deep neural network.  ...  Precup for their comments and suggestions on the paper.  ... 
arXiv:2007.08433v1 fatcat:ljl2ig64rffmphbluh24zpceoq

Unsupervised Meta-Learning for Reinforcement Learning [article]

Abhishek Gupta, Benjamin Eysenbach, Chelsea Finn, Sergey Levine
2020 arXiv   pre-print
Our conceptual and theoretical contributions consist of formulating the unsupervised meta-reinforcement learning problem and describing how task proposals based on mutual information can be used to train  ...  The performance of meta-learning algorithms depends on the tasks available for meta-training: in the same way that supervised learning generalizes best to test points drawn from the same distribution as  ...  In practice, the performance of meta-learning algorithms depends on the user-specified meta-training task distribution.  ... 
arXiv:1806.04640v3 fatcat:p4exyyqnnbesvjatymjqhiz2du

Offline Meta-Reinforcement Learning for Industrial Insertion [article]

Tony Z. Zhao, Jianlan Luo, Oleg Sushkov, Rugile Pevceviciute, Nicolas Heess, Jon Scholz, Stefan Schaal, Sergey Levine
2022 arXiv   pre-print
In this paper, we tackle rapid adaptation to new tasks through the framework of meta-learning, which utilizes past tasks to learn to adapt with a specific focus on industrial insertion tasks.  ...  In this setting, we address two specific challenges when applying meta-learning. First, conventional meta-RL algorithms require lengthy online meta-training.  ...  Offline and Online Reinforcement Learning In order to meta-train on offline data, we require a suitable offline RL algorithm.  ... 
arXiv:2110.04276v3 fatcat:illo34j54jaofkqom3kqvkzwwm

Learning to Explore via Meta-Policy Gradient

Tianbing Xu, Qiang Liu, Liang Zhao, Jian Peng
2018 International Conference on Machine Learning  
With an extensive study, we show that our method significantly improves the sample-efficiency of DDPG on a variety of reinforcement learning continuous control tasks.  ...  In this work, we develop a simple meta-policy gradient algorithm that allows us to adaptively learn the exploration policy in DDPG.  ...  Acknowledgement We sincerely appreciate the constructive comments from our three anonymous reviewers, which improve our paper significantly.  ... 
dblp:conf/icml/XuLZP18 fatcat:2yvq5wq4kzckdcdmdaqfzwweju

A Survey on Deep Reinforcement Learning-based Approaches for Adaptation and Generalization [article]

Pamul Yadav, Ashutosh Mishra, Junyong Lee, Shiho Kim
2022 arXiv   pre-print
This paper presents a survey on the recent developments in DRL-based approaches for adaptation and generalization. We begin by formulating these goals in the context of task and domain.  ...  Deep Reinforcement Learning (DRL) aims to create intelligent agents that can learn to solve complex problems efficiently in a real-world environment.  ...  Meta Reinforcement Learning (Meta-RL) Meta-RL is based on the notion of learning to learn [Thrun and Pratt, 1998 ].  ... 
arXiv:2202.08444v1 fatcat:xc3bgq3jdngazlplw66ejtax6q
« Previous Showing results 1 — 15 out of 25,057 results