Self-reflective deep reinforcement learning

Abdulrahman Altahhan

doi:10.1109/IJCNN.2016.7727798

Self-reflective deep reinforcement learning

Abdulrahman Altahhan

Research output: Contribution to conference › Paper › peer-review

2 Citations (Scopus)

73 Downloads (Pure)

Abstract

In this paper we present a new concept of self-reflection learning to support a deep reinforcement learning model. The self-reflective process occurs offline between episodes to help the agent to learn to navigate towards a goal location and boost its online performance. In particular, a so far optimal experience is recalled and compared with other similar but suboptimal episodes to reemphasize worthy decisions and deemphasize unworthy ones using eligibility and learning traces. At the same time, relatively bad experience is forgotten to remove its confusing effect. We set up a layer-wise deep actor-critic architecture and apply the self-reflection process to help to train it. We show that the self-reflective model seems to work well and initial experimental result on real robot shows that the agent accomplished good success rate in reaching a goal location.

Original language	English
Pages	4565 - 4570
DOIs	https://doi.org/10.1109/IJCNN.2016.7727798
Publication status	Published - 3 Nov 2016
Event	2016 International Joint Conference on Neural Networks - Vancouver, Canada Duration: 24 Jul 2016 → 29 Jul 2016

Conference

Conference	2016 International Joint Conference on Neural Networks
Abbreviated title	IJCNN
Country/Territory	Canada
City	Vancouver
Period	24/07/16 → 29/07/16

Bibliographical note

© 2016 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works

Keywords

robot navigation
self-reflective deep reinforcement learning
deep learning
actor-critic
neural networks

Access to Document

10.1109/IJCNN.2016.7727798

WCCI 2016 self reflective learning 3 submitted-amended (2)

Cite this

@conference{f44984bf9d6e4caf97e2701a61a2d800,

title = "Self-reflective deep reinforcement learning",

abstract = "In this paper we present a new concept of self-reflection learning to support a deep reinforcement learning model. The self-reflective process occurs offline between episodes to help the agent to learn to navigate towards a goal location and boost its online performance. In particular, a so far optimal experience is recalled and compared with other similar but suboptimal episodes to reemphasize worthy decisions and deemphasize unworthy ones using eligibility and learning traces. At the same time, relatively bad experience is forgotten to remove its confusing effect. We set up a layer-wise deep actor-critic architecture and apply the self-reflection process to help to train it. We show that the self-reflective model seems to work well and initial experimental result on real robot shows that the agent accomplished good success rate in reaching a goal location. ",

keywords = "robot navigation, self-reflective deep reinforcement learning, deep learning, actor-critic, neural networks",

author = "Abdulrahman Altahhan",

note = "{\textcopyright} 2016 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works; 2016 International Joint Conference on Neural Networks, IJCNN ; Conference date: 24-07-2016 Through 29-07-2016",

year = "2016",

month = nov,

day = "3",

doi = "10.1109/IJCNN.2016.7727798",

language = "English",

pages = "4565 -- 4570",

}

TY - CONF

T1 - Self-reflective deep reinforcement learning

AU - Altahhan, Abdulrahman

N1 - © 2016 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works

PY - 2016/11/3

Y1 - 2016/11/3

N2 - In this paper we present a new concept of self-reflection learning to support a deep reinforcement learning model. The self-reflective process occurs offline between episodes to help the agent to learn to navigate towards a goal location and boost its online performance. In particular, a so far optimal experience is recalled and compared with other similar but suboptimal episodes to reemphasize worthy decisions and deemphasize unworthy ones using eligibility and learning traces. At the same time, relatively bad experience is forgotten to remove its confusing effect. We set up a layer-wise deep actor-critic architecture and apply the self-reflection process to help to train it. We show that the self-reflective model seems to work well and initial experimental result on real robot shows that the agent accomplished good success rate in reaching a goal location.

AB - In this paper we present a new concept of self-reflection learning to support a deep reinforcement learning model. The self-reflective process occurs offline between episodes to help the agent to learn to navigate towards a goal location and boost its online performance. In particular, a so far optimal experience is recalled and compared with other similar but suboptimal episodes to reemphasize worthy decisions and deemphasize unworthy ones using eligibility and learning traces. At the same time, relatively bad experience is forgotten to remove its confusing effect. We set up a layer-wise deep actor-critic architecture and apply the self-reflection process to help to train it. We show that the self-reflective model seems to work well and initial experimental result on real robot shows that the agent accomplished good success rate in reaching a goal location.

KW - robot navigation

KW - self-reflective deep reinforcement learning

KW - deep learning

KW - actor-critic

KW - neural networks

U2 - 10.1109/IJCNN.2016.7727798

DO - 10.1109/IJCNN.2016.7727798

M3 - Paper

SP - 4565

EP - 4570

T2 - 2016 International Joint Conference on Neural Networks

Y2 - 24 July 2016 through 29 July 2016

ER -

Self-reflective deep reinforcement learning

Abstract

Conference

Bibliographical note

Keywords

Access to Document

Fingerprint

Cite this