Countering poisonous inputs with memetic neuroevolution

Julian Togelius; Tom Schaul; Jürgen Schmidhuber; Faustino Gomez

doi:10.1007/978-3-540-87700-4_61

Countering poisonous inputs with memetic neuroevolution

Julian Togelius, Tom Schaul, Jürgen Schmidhuber, Faustino Gomez

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

6 Scopus citations

Abstract

Applied to certain problems, neuroevolution frequently gets stuck in local optima with very low fitness; in particular, this is true for some reinforcement learning problems where the input to the controller is a high-dimensional and/or ill-chosen state description. Evidently, some controller inputs are "poisonous", and their inclusion induce such local optima. Previously, we proposed the memetic climber, which evolves neural network topology and weights at different timescales, as a solution to this problem. In this paper, we further explore the memetic climber, and introduce its population-based counterpart: the memetic ES. We also explore which types of inputs are poisonous for two different reinforcement learning problems. © 2008 Springer-Verlag Berlin Heidelberg.

Original language	English (US)
Title of host publication	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Pages	610-619
Number of pages	10
DOIs	https://doi.org/10.1007/978-3-540-87700-4_61
State	Published - Nov 26 2008
Externally published	Yes

ASJC Scopus subject areas

Theoretical Computer Science
General Computer Science

Access to Document

10.1007/978-3-540-87700-4_61

Cite this

@inproceedings{4509345de4d74ad999c95f40ed0aa5be,

title = "Countering poisonous inputs with memetic neuroevolution",

abstract = "Applied to certain problems, neuroevolution frequently gets stuck in local optima with very low fitness; in particular, this is true for some reinforcement learning problems where the input to the controller is a high-dimensional and/or ill-chosen state description. Evidently, some controller inputs are {"}poisonous{"}, and their inclusion induce such local optima. Previously, we proposed the memetic climber, which evolves neural network topology and weights at different timescales, as a solution to this problem. In this paper, we further explore the memetic climber, and introduce its population-based counterpart: the memetic ES. We also explore which types of inputs are poisonous for two different reinforcement learning problems. {\textcopyright} 2008 Springer-Verlag Berlin Heidelberg.",

author = "Julian Togelius and Tom Schaul and J{\"u}rgen Schmidhuber and Faustino Gomez",

note = "Generated from Scopus record by KAUST IRTS on 2022-09-14",

year = "2008",

month = nov,

day = "26",

doi = "10.1007/978-3-540-87700-4_61",

language = "English (US)",

isbn = "3540876995",

pages = "610--619",

booktitle = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

}

TY - GEN

T1 - Countering poisonous inputs with memetic neuroevolution

AU - Togelius, Julian

AU - Schaul, Tom

AU - Schmidhuber, Jürgen

AU - Gomez, Faustino

N1 - Generated from Scopus record by KAUST IRTS on 2022-09-14

PY - 2008/11/26

Y1 - 2008/11/26

N2 - Applied to certain problems, neuroevolution frequently gets stuck in local optima with very low fitness; in particular, this is true for some reinforcement learning problems where the input to the controller is a high-dimensional and/or ill-chosen state description. Evidently, some controller inputs are "poisonous", and their inclusion induce such local optima. Previously, we proposed the memetic climber, which evolves neural network topology and weights at different timescales, as a solution to this problem. In this paper, we further explore the memetic climber, and introduce its population-based counterpart: the memetic ES. We also explore which types of inputs are poisonous for two different reinforcement learning problems. © 2008 Springer-Verlag Berlin Heidelberg.

AB - Applied to certain problems, neuroevolution frequently gets stuck in local optima with very low fitness; in particular, this is true for some reinforcement learning problems where the input to the controller is a high-dimensional and/or ill-chosen state description. Evidently, some controller inputs are "poisonous", and their inclusion induce such local optima. Previously, we proposed the memetic climber, which evolves neural network topology and weights at different timescales, as a solution to this problem. In this paper, we further explore the memetic climber, and introduce its population-based counterpart: the memetic ES. We also explore which types of inputs are poisonous for two different reinforcement learning problems. © 2008 Springer-Verlag Berlin Heidelberg.

UR - http://link.springer.com/10.1007/978-3-540-87700-4_61

UR - http://www.scopus.com/inward/record.url?scp=56449113460&partnerID=8YFLogxK

U2 - 10.1007/978-3-540-87700-4_61

DO - 10.1007/978-3-540-87700-4_61

M3 - Conference contribution

SN - 3540876995

SP - 610

EP - 619

BT - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

ER -

Countering poisonous inputs with memetic neuroevolution

Abstract

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this