Fast Learning with Predictive Forward Models

Carlos Brody

Fast Learning with Predictive Forward Models

Carlos Brody

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

A method for transforming performance evaluation signals distal both in space and time into proximal signals usable by supervised learning algorithms, presented in [Jordan ii Jacobs 90], is examined. A simple observation concerning differentiation through models trained with redundant inputs (as one of their networks is) explains a weakness in the original architecture and suggests a modification: an internal world model that encodes action-space exploration and, crucially, cancels input redundancy to the forward model is added. Learning time on an example task, cart-pole balancing, is thereby reduced about 50 to 100 times.

Original language	English (US)
Title of host publication	Advances in Neural Information Processing Systems 4, NIPS 1991
Editors	John E. Moody, Stephen Jose Hanson, Richard Lippmann
Publisher	Neural information processing systems foundation
Pages	563-570
Number of pages	8
ISBN (Electronic)	1558602224, 9781558602229
State	Published - 1991
Externally published	Yes
Event	4th Advances in Neural Information Processing Systems, NIPS 1991 - Denver, United States Duration: Dec 2 1991 → Dec 5 1991

Publication series

Name	Advances in Neural Information Processing Systems
Volume	4
ISSN (Print)	1049-5258

Conference

Conference	4th Advances in Neural Information Processing Systems, NIPS 1991
Country/Territory	United States
City	Denver
Period	12/2/91 → 12/5/91

All Science Journal Classification (ASJC) codes

Signal Processing
Information Systems
Computer Networks and Communications

Cite this

@inproceedings{121fb588800041558cf547c045dc8070,

title = "Fast Learning with Predictive Forward Models",

abstract = "A method for transforming performance evaluation signals distal both in space and time into proximal signals usable by supervised learning algorithms, presented in [Jordan ii Jacobs 90], is examined. A simple observation concerning differentiation through models trained with redundant inputs (as one of their networks is) explains a weakness in the original architecture and suggests a modification: an internal world model that encodes action-space exploration and, crucially, cancels input redundancy to the forward model is added. Learning time on an example task, cart-pole balancing, is thereby reduced about 50 to 100 times.",

author = "Carlos Brody",

note = "Publisher Copyright: {\textcopyright} 1991 Neural information processing systems foundation. All rights reserved.; 4th Advances in Neural Information Processing Systems, NIPS 1991 ; Conference date: 02-12-1991 Through 05-12-1991",

year = "1991",

language = "English (US)",

series = "Advances in Neural Information Processing Systems",

publisher = "Neural information processing systems foundation",

pages = "563--570",

editor = "Moody, \{John E.\} and Hanson, \{Stephen Jose\} and Richard Lippmann",

booktitle = "Advances in Neural Information Processing Systems 4, NIPS 1991",

}

Brody, C 1991, Fast Learning with Predictive Forward Models. in JE Moody, SJ Hanson & R Lippmann (eds), Advances in Neural Information Processing Systems 4, NIPS 1991. Advances in Neural Information Processing Systems, vol. 4, Neural information processing systems foundation, pp. 563-570, 4th Advances in Neural Information Processing Systems, NIPS 1991, Denver, United States, 12/2/91.

Fast Learning with Predictive Forward Models. / Brody, Carlos.
Advances in Neural Information Processing Systems 4, NIPS 1991. ed. / John E. Moody; Stephen Jose Hanson; Richard Lippmann. Neural information processing systems foundation, 1991. p. 563-570 (Advances in Neural Information Processing Systems; Vol. 4).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Fast Learning with Predictive Forward Models

AU - Brody, Carlos

PY - 1991

Y1 - 1991

N2 - A method for transforming performance evaluation signals distal both in space and time into proximal signals usable by supervised learning algorithms, presented in [Jordan ii Jacobs 90], is examined. A simple observation concerning differentiation through models trained with redundant inputs (as one of their networks is) explains a weakness in the original architecture and suggests a modification: an internal world model that encodes action-space exploration and, crucially, cancels input redundancy to the forward model is added. Learning time on an example task, cart-pole balancing, is thereby reduced about 50 to 100 times.

AB - A method for transforming performance evaluation signals distal both in space and time into proximal signals usable by supervised learning algorithms, presented in [Jordan ii Jacobs 90], is examined. A simple observation concerning differentiation through models trained with redundant inputs (as one of their networks is) explains a weakness in the original architecture and suggests a modification: an internal world model that encodes action-space exploration and, crucially, cancels input redundancy to the forward model is added. Learning time on an example task, cart-pole balancing, is thereby reduced about 50 to 100 times.

UR - https://www.scopus.com/pages/publications/105021104393

UR - https://www.scopus.com/pages/publications/105021104393#tab=citedBy

M3 - Conference contribution

AN - SCOPUS:105021104393

T3 - Advances in Neural Information Processing Systems

SP - 563

EP - 570

BT - Advances in Neural Information Processing Systems 4, NIPS 1991

A2 - Moody, John E.

A2 - Hanson, Stephen Jose

A2 - Lippmann, Richard

PB - Neural information processing systems foundation

T2 - 4th Advances in Neural Information Processing Systems, NIPS 1991

Y2 - 2 December 1991 through 5 December 1991

ER -

Fast Learning with Predictive Forward Models

Abstract

Publication series

Conference

All Science Journal Classification (ASJC) codes

Other files and links

Fingerprint

Cite this