Programmatic Reinforcement Learning using Critic-Moderated Evolution

Senne Deproost; Denis Steckelmacher; Ann Nowe

Programmatic Reinforcement Learning using Critic-Moderated Evolution

Senne Deproost, Denis Steckelmacher, Ann Nowe

Research output: Unpublished contribution to conference › Poster

16 Downloads (Pure)

Abstract

We propose a new method to generate a program from a Reinforcement Learning policy. Compared to previous methods, we exploit more RL-specific elements such as the critic value-network. Improved actions from the critic are used to steer a Genetic Programming process via a fitness function.

Original language	English
Publication status	Unpublished - 18 Nov 2024
Event	BNAIC/BeNeLearn 2024: Joint International Scientific Conferences on AI and Machine Learning - Jaarbeurs Supernova, Utrecht, Netherlands Duration: 18 Nov 2024 → 20 Nov 2024 Conference number: 36 https://bnaic2024.sites.uu.nl/ https://bnaic2024.sites.uu.nl

Conference

Conference	BNAIC/BeNeLearn 2024: Joint International Scientific Conferences on AI and Machine Learning
Abbreviated title	BNAIC/BeNeLearn 2024
Country/Territory	Netherlands
City	Utrecht
Period	18/11/24 → 20/11/24
Internet address	https://bnaic2024.sites.uu.nl/ https://bnaic2024.sites.uu.nl

Keywords

Deep Reinforcement Learning
Genetic Programming
Explainable AI

Access to Document

BNAIC2024_posterFinal published version, 235 KBLicence: CC BY

2 Active

VLAAI1: Flanders Artificial Intelligence Research program (FAIR) – second cycle
Nowe, A. & Vanderborght, B.
1/01/24 → 31/12/28
Project: Applied
FWOSBO46: SBO Project: CTRL schemes merged with eXplainable AI for t(h)rustworthy control of physical dynamic systems (CTRLxAI=T(H)RUST)
Nowe, A., Steckelmacher, D., Efthymiadis, K. & Schietgat, L.
1/10/22 → 30/09/26
Project: Applied

1 Conference paper

Human-readable programs as the actor of a Reinforcement Learning agent using Critic-Moderated Evolution
Deproost, S., Steckelmacher, D. & Nowe, A., 17 Nov 2024, Proceedings of the 2024 Benelux Conference on Artificial Intelligence. Benelux Association for Artificial Intelligence (BNVKI-AIABN), p. 1-18 14 p. (BNAIC Proceedings).
Research output: Chapter in Book/Report/Conference proceeding › Conference paper › Research

Open Access

2 Participation in conference
1 Talk or presentation at a conference

Flanders AI Internal Research Day
Senne Deproost (Participant)
10 Mar 2025
Activity: Participating in or organising an event › Participation in conference
Critic-Moderated Genetic Programming
Senne Deproost (Presenter)
18 Nov 2024
Activity: Talk or presentation › Talk or presentation at a conference
BNAIC/BeNeLearn 2024: Joint International Scientific Conferences on AI and Machine Learning
Senne Deproost (Participant)
18 Nov 2024 → 20 Nov 2024
Activity: Participating in or organising an event › Participation in conference

Best Paper Award
Deproost, Senne (Recipient), Steckelmacher, Denis (Recipient) & Nowe, Ann (Recipient), 20 Nov 2024
Prize: Prize (including medals and awards)

Cite this

@conference{9b2db174ac0641ba9bec1057285a292d,

title = "Programmatic Reinforcement Learning using Critic-Moderated Evolution",

abstract = "We propose a new method to generate a program from a Reinforcement Learning policy. Compared to previous methods, we exploit more RL-specific elements such as the critic value-network. Improved actions from the critic are used to steer a Genetic Programming process via a fitness function.",

keywords = "Deep Reinforcement Learning, Genetic Programming, Explainable AI",

author = "Senne Deproost and Denis Steckelmacher and Ann Nowe",

year = "2024",

month = nov,

day = "18",

language = "English",

note = "BNAIC/BeNeLearn 2024: Joint International Scientific Conferences on AI and Machine Learning, BNAIC/BeNeLearn 2024 ; Conference date: 18-11-2024 Through 20-11-2024",

url = "https://bnaic2024.sites.uu.nl/, https://bnaic2024.sites.uu.nl",

}

TY - CONF

T1 - Programmatic Reinforcement Learning using Critic-Moderated Evolution

AU - Deproost, Senne

AU - Steckelmacher, Denis

AU - Nowe, Ann

N1 - Conference code: 36

PY - 2024/11/18

Y1 - 2024/11/18

N2 - We propose a new method to generate a program from a Reinforcement Learning policy. Compared to previous methods, we exploit more RL-specific elements such as the critic value-network. Improved actions from the critic are used to steer a Genetic Programming process via a fitness function.

AB - We propose a new method to generate a program from a Reinforcement Learning policy. Compared to previous methods, we exploit more RL-specific elements such as the critic value-network. Improved actions from the critic are used to steer a Genetic Programming process via a fitness function.

KW - Deep Reinforcement Learning

KW - Genetic Programming

KW - Explainable AI

M3 - Poster

T2 - BNAIC/BeNeLearn 2024: Joint International Scientific Conferences on AI and Machine Learning

Y2 - 18 November 2024 through 20 November 2024

ER -

Programmatic Reinforcement Learning using Critic-Moderated Evolution

Abstract

Conference

Keywords

Access to Document

Projects

VLAAI1: Flanders Artificial Intelligence Research program (FAIR) – second cycle

FWOSBO46: SBO Project: CTRL schemes merged with eXplainable AI for t(h)rustworthy control of physical dynamic systems (CTRLxAI=T(H)RUST)

Research output

Human-readable programs as the actor of a Reinforcement Learning agent using Critic-Moderated Evolution

Activities

Flanders AI Internal Research Day

Critic-Moderated Genetic Programming

BNAIC/BeNeLearn 2024: Joint International Scientific Conferences on AI and Machine Learning

Prizes

Best Paper Award

Cite this