Online Optimization of Curriculum Learning Schedules using Evolutionary Optimization

Publikation: Beitrag in Buch/Bericht/Sammelwerk/KonferenzbandAufsatz in KonferenzbandForschungPeer-Review

Autorschaft

Forschungs-netzwerk anzeigen

Details

OriginalspracheEnglisch
Titel des SammelwerksProceedings of the 2024 IEEE Conference on Games, CoG 2024
Herausgeber (Verlag)IEEE Computer Society
ISBN (elektronisch)9798350350678
ISBN (Print)979-8-3503-5068-5
PublikationsstatusVeröffentlicht - 5 Aug. 2024
Veranstaltung6th Annual IEEE Conference on Games, CoG 2024 - Milan, Italien
Dauer: 5 Aug. 20248 Aug. 2024

Publikationsreihe

NameIEEE Conference on Computatonal Intelligence and Games, CIG
ISSN (Print)2325-4270
ISSN (elektronisch)2325-4289

Abstract

We propose RHEA CL, which combines Curriculum Learning (CL) with Rolling Horizon Evolutionary Algorithms (RHEA) to automatically produce effective curricula during the training of a reinforcement learning agent. RHEA CL optimizes a population of curricula, using an evolutionary algorithm, and selects the best-performing curriculum as the starting point for the next training epoch. Performance evaluations are conducted after every curriculum step in all environments. We evaluate the algorithm on the DoorKey and DynamicObstacles environments within the Minigrid framework. It demonstrates adaptability and consistent improvement, particularly in the early stages, while reaching a stable performance later that is capable of outperforming other curriculum learners. In comparison to other curriculum schedules, RHEA CL has shown to yield performance improvements for the final Reinforcement learning (RL) agent at the cost of additional evaluation during training.

ASJC Scopus Sachgebiete

Zitieren

Online Optimization of Curriculum Learning Schedules using Evolutionary Optimization. / Jiwatode, Mohit; Schlecht, Leon; Dockhorn, Alexander.
Proceedings of the 2024 IEEE Conference on Games, CoG 2024. IEEE Computer Society, 2024. (IEEE Conference on Computatonal Intelligence and Games, CIG).

Publikation: Beitrag in Buch/Bericht/Sammelwerk/KonferenzbandAufsatz in KonferenzbandForschungPeer-Review

Jiwatode, M, Schlecht, L & Dockhorn, A 2024, Online Optimization of Curriculum Learning Schedules using Evolutionary Optimization. in Proceedings of the 2024 IEEE Conference on Games, CoG 2024. IEEE Conference on Computatonal Intelligence and Games, CIG, IEEE Computer Society, 6th Annual IEEE Conference on Games, CoG 2024, Milan, Italien, 5 Aug. 2024. https://doi.org/10.48550/arXiv.2408.06068, https://doi.org/10.1109/CoG60054.2024.10645570
Jiwatode, M., Schlecht, L., & Dockhorn, A. (2024). Online Optimization of Curriculum Learning Schedules using Evolutionary Optimization. In Proceedings of the 2024 IEEE Conference on Games, CoG 2024 (IEEE Conference on Computatonal Intelligence and Games, CIG). IEEE Computer Society. https://doi.org/10.48550/arXiv.2408.06068, https://doi.org/10.1109/CoG60054.2024.10645570
Jiwatode M, Schlecht L, Dockhorn A. Online Optimization of Curriculum Learning Schedules using Evolutionary Optimization. in Proceedings of the 2024 IEEE Conference on Games, CoG 2024. IEEE Computer Society. 2024. (IEEE Conference on Computatonal Intelligence and Games, CIG). doi: 10.48550/arXiv.2408.06068, 10.1109/CoG60054.2024.10645570
Jiwatode, Mohit ; Schlecht, Leon ; Dockhorn, Alexander. / Online Optimization of Curriculum Learning Schedules using Evolutionary Optimization. Proceedings of the 2024 IEEE Conference on Games, CoG 2024. IEEE Computer Society, 2024. (IEEE Conference on Computatonal Intelligence and Games, CIG).
Download
@inproceedings{a2d3981b66c8426aa66e811e2679164a,
title = "Online Optimization of Curriculum Learning Schedules using Evolutionary Optimization",
abstract = "We propose RHEA CL, which combines Curriculum Learning (CL) with Rolling Horizon Evolutionary Algorithms (RHEA) to automatically produce effective curricula during the training of a reinforcement learning agent. RHEA CL optimizes a population of curricula, using an evolutionary algorithm, and selects the best-performing curriculum as the starting point for the next training epoch. Performance evaluations are conducted after every curriculum step in all environments. We evaluate the algorithm on the DoorKey and DynamicObstacles environments within the Minigrid framework. It demonstrates adaptability and consistent improvement, particularly in the early stages, while reaching a stable performance later that is capable of outperforming other curriculum learners. In comparison to other curriculum schedules, RHEA CL has shown to yield performance improvements for the final Reinforcement learning (RL) agent at the cost of additional evaluation during training.",
keywords = "Curriculum Learning, Evolutionary Algorithms, Reinforcement Learning, Rolling Horizon Algorithms",
author = "Mohit Jiwatode and Leon Schlecht and Alexander Dockhorn",
note = "Publisher Copyright: {\textcopyright} 2024 IEEE.; 6th Annual IEEE Conference on Games, CoG 2024 ; Conference date: 05-08-2024 Through 08-08-2024",
year = "2024",
month = aug,
day = "5",
doi = "10.48550/arXiv.2408.06068",
language = "English",
isbn = "979-8-3503-5068-5",
series = "IEEE Conference on Computatonal Intelligence and Games, CIG",
publisher = "IEEE Computer Society",
booktitle = "Proceedings of the 2024 IEEE Conference on Games, CoG 2024",
address = "United States",

}

Download

TY - GEN

T1 - Online Optimization of Curriculum Learning Schedules using Evolutionary Optimization

AU - Jiwatode, Mohit

AU - Schlecht, Leon

AU - Dockhorn, Alexander

N1 - Publisher Copyright: © 2024 IEEE.

PY - 2024/8/5

Y1 - 2024/8/5

N2 - We propose RHEA CL, which combines Curriculum Learning (CL) with Rolling Horizon Evolutionary Algorithms (RHEA) to automatically produce effective curricula during the training of a reinforcement learning agent. RHEA CL optimizes a population of curricula, using an evolutionary algorithm, and selects the best-performing curriculum as the starting point for the next training epoch. Performance evaluations are conducted after every curriculum step in all environments. We evaluate the algorithm on the DoorKey and DynamicObstacles environments within the Minigrid framework. It demonstrates adaptability and consistent improvement, particularly in the early stages, while reaching a stable performance later that is capable of outperforming other curriculum learners. In comparison to other curriculum schedules, RHEA CL has shown to yield performance improvements for the final Reinforcement learning (RL) agent at the cost of additional evaluation during training.

AB - We propose RHEA CL, which combines Curriculum Learning (CL) with Rolling Horizon Evolutionary Algorithms (RHEA) to automatically produce effective curricula during the training of a reinforcement learning agent. RHEA CL optimizes a population of curricula, using an evolutionary algorithm, and selects the best-performing curriculum as the starting point for the next training epoch. Performance evaluations are conducted after every curriculum step in all environments. We evaluate the algorithm on the DoorKey and DynamicObstacles environments within the Minigrid framework. It demonstrates adaptability and consistent improvement, particularly in the early stages, while reaching a stable performance later that is capable of outperforming other curriculum learners. In comparison to other curriculum schedules, RHEA CL has shown to yield performance improvements for the final Reinforcement learning (RL) agent at the cost of additional evaluation during training.

KW - Curriculum Learning

KW - Evolutionary Algorithms

KW - Reinforcement Learning

KW - Rolling Horizon Algorithms

UR - http://www.scopus.com/inward/record.url?scp=85203509151&partnerID=8YFLogxK

U2 - 10.48550/arXiv.2408.06068

DO - 10.48550/arXiv.2408.06068

M3 - Conference contribution

AN - SCOPUS:85203509151

SN - 979-8-3503-5068-5

T3 - IEEE Conference on Computatonal Intelligence and Games, CIG

BT - Proceedings of the 2024 IEEE Conference on Games, CoG 2024

PB - IEEE Computer Society

T2 - 6th Annual IEEE Conference on Games, CoG 2024

Y2 - 5 August 2024 through 8 August 2024

ER -

Von denselben Autoren