Recherche doctorale au LIST · financée par le FNR

De meilleurs films polymères en moins d’expériences.

POLYSUN fabrique des films de polymères conjugués en phase vapeur. Un modèle d’IA choisit la prochaine expérience par optimisation bayésienne.

Touchez la carte pour lancer une expérience. Trouverez-vous le sommet ?

Le projet

Une chimie qui apprend de chaque essai

Moins d’essais : un modèle d’IA apprend de chaque résultat et propose l’expérience suivante.

Les polymères conjugués conduisent l’électricité et interagissent avec la lumière, ce qui les rend utiles en couches minces pour l’électronique, les capteurs et l’électrochimie. POLYSUN les fabrique par dépôt chimique en phase vapeur oxydant (oCVD) : le monomère et un oxydant sont évaporés sous vide et réagissent directement sur la surface, sans solvant.

Les propriétés du film dépendent de nombreuses conditions à la fois, comme les températures et les rapports de réactifs. Plutôt que de les tester une par une, POLYSUN utilise l’optimisation bayésienne par régression de processus gaussiens : après chaque expérience, un modèle statistique est mis à jour et propose l’expérience suivante la plus informative.

Le projet automatise aussi l’installation d’oCVD, afin que l’optimiseur, l’instrument et les mesures fonctionnent ensemble en boucle fermée.

Fabriquer

Des films polymères déposés en phase vapeur par oCVD, sans solvant.

Apprendre

Un modèle de processus gaussien se met à jour après chaque expérience.

Décider

L’optimisation bayésienne choisit l’expérience suivante.

Envie de voir ce que fait l’optimiseur ? Essayez vous-même.

Le jeu

Pouvez-vous battre l’optimiseur ?

Trouvez les meilleures conditions d’une réaction fictive, puis regardez un modèle d’IA (un optimiseur bayésien) faire de même. Le moins d’expériences gagne.

Optimize This!

Jeu éducatif gratuit. Sans compte, sans publicité, dans votre navigateur.

2formats
5modificateurs
0compte ou publicité
▶ Jouer dans le navigateur

Le chercheur

Jonathan Hadji-Minaglou

Doctorant au LIST et à l’Université du Luxembourg

Je suis chimiste et je travaille à la croisée de la science des matériaux et de l’apprentissage automatique. Dans POLYSUN, lancé en septembre 2025, je fabrique des films de polymères conjugués par dépôt chimique en phase vapeur oxydant (oCVD) et j’utilise l’optimisation bayésienne par régression de processus gaussiens pour choisir l’expérience suivante. J’ai aussi automatisé l’installation d’oCVD et écrit le logiciel qui pilote ces campagnes d’optimisation.

En dehors du labo, j’ai créé Optimize This! pour montrer comment fonctionne l’optimisation bayésienne, et je présente le projet lors de conférences et d’écoles d’été, avec notamment des prix de poster à HYMAT et à l’école d’été AiMat (KIT).

Encadrant Matthias Rupp Chef du groupe Process modelling, automation and robotisation (PROMAR) au LIST
Me contacter
Qu’en est-il sorti jusqu’ici ?

Publications

Premiers articles en préparation

    Les articles évalués par les pairs, les prépublications et les jeux de données de POLYSUN seront listés ici dès leur publication.

    Qui le rend possible ?

    Financement

    Soutenu par le FNR

    Le projet POLYSUN est soutenu par le Luxembourg National Research Fund (FNR) dans le cadre de son programme AFR pour la recherche doctorale.

    Référence FNR AFR 19478064

    La recherche est menée au Luxembourg Institute of Science and Technology (LIST), dans le cadre d’un doctorat à l’Université du Luxembourg.

    Luxembourg National Research Fund (FNR) Luxembourg Institute of Science and Technology (LIST) University of Luxembourg

    Les textes juridiques sont disponibles en anglais.

    Web site editor

    This web site is edited by the Luxembourg Institute of Science and Technology (LIST). For any question concerning this web site or its content, do not hesitate to contact us:

    Luxembourg Institute of Science and Technology
    Attn: Jonathan HADJI-MINAGLOU
    5, avenue des Hauts Fourneaux
    L-4362 Esch-sur-Alzette
    Luxembourg

    Email: jonathan.minaglou [at] list [dot] lu
    RCS number: J53
    VAT number: LU 27327040

    Purpose

    The purpose of this website is to present and make available the educational application “Optimize This!” developed by LIST in the framework of the “POLYSUN” research project.

    Funding

    The project “POLYSUN” is supported by the National Research Fund, Luxembourg.

    Luxembourg National Research Fund (FNR)

    Use of this website

    LIST accepts no responsibility with regard to:

    This web site may provide links to sites that are not under LIST’s control. Therefore, we accept no responsibility with any of their content, products or services.

    The present liability section is not intended to limit any requirements laid down in applicable law or to exclude liability for matters which may not be excluded under applicable law.

    Intellectual Property

    This website (including any text, image, video or software) is protected by Intellectual Property, including copyright, trademarks and patents. LIST (and its licensors) reserves all these rights.

    © 2026, all rights reserved.

    Modifications

    We may change this legal notice at any time and without any prior notice.

    Applicable law and jurisdiction

    This legal notice and the use of this website shall be governed by the law of Luxembourg and any related litigation subject to the exclusive Luxembourg jurisdiction.

    LIST’s privacy notice applies to this website: www.list.lu/privacy-notice. LIST’s general legal notice: www.list.lu/legal-notice.

    Terms of use – Optimize This!

    Les textes juridiques sont disponibles en anglais.

    Version 1: 4 November 2026

    1. Description of the application

    Optimize This! is an educational application (hereafter the “Application”) that helps a wide audience understand how Bayesian Optimization using Gaussian Process Regression works, and develop an intuition for it, by competing against the optimizer.

    Each game starts with a fictive scenario chosen randomly by the Application. It sets out your role, your project, the parameters to optimize, the price per experiment and the time per experiment. The goal is to maximize the yield of a fictive chemical reaction in as few experiments as possible.

    In the standard format, the human player first optimizes the reaction by tweaking the reaction parameters. The Bayesian Optimizer (Matérn 5/2 kernel, Expected Improvement acquisition function) then optimizes the same reaction. In the alternative “Alternating turns” format, the player and the Optimizer take turns on the same landscape, and the first to reach the target wins. The winner is the one who optimizes the reaction in the fewest steps.

    Optional modifiers are:

    After each game you can play again and find supplementary information on the end screen about how Bayesian Optimization via Gaussian Process Regression works. Modes and features may change over time.

    Optimize This! was developed by the Luxembourg Institute of Science and Technology (LIST) as part of the project “POLYSUN”, supported by the National Research Fund, Luxembourg.

    2. Educational and fictional nature

    Scenarios, reactions, parameters, costs, durations, yields and purity values are fictive or generated for gameplay. They are not experimental results or recommendations. Do not use them for safety, engineering, commercial or scientific decisions about real experiments or chemical processes. The Application is a simplified teaching tool and does not claim to be a complete or universally applicable account of Bayesian Optimization.

    The Application is intended for use by adults, including teachers and other educators. Teachers, schools and other educational organisations that make the Application available to minors are responsible for appropriate supervision of that use.

    3. Permitted use

    By accessing or using the Application, you accept these terms and commit in using the Application in accordance with them as well as all applicable laws and regulations, and any rules or instructions displayed in the Application. In particular, you must not:

    4. Access

    The Application is made available free of charge and no login is required.

    LIST may modify, suspend or discontinue all or part of the application, including for maintenance, security, technical or legal reasons.

    5. Intellectual property and copyright

    © 2026 Luxembourg Institute of Science and Technology (LIST). All rights reserved in LIST’s own code and content. Except as permitted by applicable law or by LIST in writing, copying or redistributing them is prohibited. Third-party components are the property of their respective owners and are used under their own licences (see section 6).

    Subject to these terms, LIST grants you a non-exclusive, non-transferable, revocable right to access and use the Application for personal, educational and non-commercial purposes. You acquire no ownership or other intellectual property rights. Without LIST’s prior written permission, you may not copy, redistribute, publicly communicate, sell or create derivative works from LIST’s own code and content.

    The name “LIST” and its logo are the property of LIST and may not be used without permission.

    6. Third-party components

    All components distributed with the software are under permissive, non-copyleft licences (MIT, ISC, 0BSD, BSD-3-Clause and Apache-2.0). They remain subject to their own licences. Copyright notices and licence texts are available in the Application under Open-source licences.

    7. Data protection

    LIST does not collect or process personal data through the Application, and the Application does not use advertising, analytics or tracking tools. If you contact LIST by email, the information you send is processed only to answer your request. Personal data may also be processed when you visit this website. Further information is available in the Legal Notice.

    8. Disclaimer and liability

    The Application is provided “as is” and “as available”, without warranty of any kind, express or implied, including accuracy, completeness, merchantability, fitness for a particular purpose and non-infringement. LIST does not guarantee uninterrupted or error-free operation. It may suspend, modify or discontinue all or part of the Application at any time, including for maintenance, security, technical or legal reasons, and may block access in case of breach of these terms. LIST is not liable for failures or delays caused by events beyond its reasonable control.

    To the maximum extent permitted by law, LIST is not liable for any indirect, special, incidental or consequential damages, or for loss of data, arising from the use of or inability to use the Application.

    9. Changes to these terms

    The current version is the version published on this website. Each amended version replaces the previous one from its effective date and applies to your use of the Application from that date. Previous versions are available on request. You are invited to consult these Terms regularly before using the Application.

    10. Applicable law and jurisdiction

    These terms are governed by Luxembourg law. The courts of Luxembourg have exclusive jurisdiction over disputes arising from these terms or the use of the Application.

    11. Contact

    For any questions about the Application or these terms please contact: jonathan.minaglou [at] list [dot] lu.

    Terms of use – Optimize This! Android app

    Les textes juridiques sont disponibles en anglais.

    Version 1: 4 November 2026

    1. Description of the application

    Optimize This! is an educational application (hereafter the “Application”) that helps a wide audience understand how Bayesian Optimization using Gaussian Process Regression works, and develop an intuition for it, by competing against the optimizer.

    Each game starts with a fictive scenario chosen randomly by the Application. It sets out your role, your project, the parameters to optimize, the price per experiment and the time per experiment. The goal is to maximize the yield of a fictive chemical reaction in as few experiments as possible.

    In the standard format, the human player first optimizes the reaction by tweaking the reaction parameters. The Bayesian Optimizer (Matérn 5/2 kernel, Expected Improvement acquisition function) then optimizes the same reaction. In the alternative “Alternating turns” format, the player and the Optimizer take turns on the same landscape, and the first to reach the target wins. The winner is the one who optimizes the reaction in the fewest steps.

    Optional modifiers are:

    After each game you can play again and find supplementary information on the end screen about how Bayesian Optimization via Gaussian Process Regression works. Modes and features may change over time.

    Optimize This! was developed by the Luxembourg Institute of Science and Technology (LIST) as part of the project “POLYSUN”, supported by the National Research Fund, Luxembourg.

    2. Educational and fictional nature

    Scenarios, reactions, parameters, costs, durations, yields and purity values are fictive or generated for gameplay. They are not experimental results or recommendations. Do not use them for safety, engineering, commercial or scientific decisions about real experiments or chemical processes. The Application is a simplified teaching tool and does not claim to be a complete or universally applicable account of Bayesian Optimization.

    The Application is intended for use by adults, including teachers and other educators. Teachers, schools and other educational organisations that make the Application available to minors are responsible for appropriate supervision of that use.

    3. Permitted use

    By accessing or using the Application, you accept these terms and commit in using the Application in accordance with them as well as all applicable laws and regulations, and any rules or instructions displayed in the Application. In particular, you must not:

    4. Access

    The Application is made available free of charge and no login is required.

    LIST may modify, suspend or discontinue all or part of the application, including for maintenance, security, technical or legal reasons.

    5. Intellectual property and copyright

    © 2026 Luxembourg Institute of Science and Technology (LIST). All rights reserved in LIST’s own code and content. Except as permitted by applicable law or by LIST in writing, copying or redistributing them is prohibited. Third-party components are the property of their respective owners and are used under their own licences (see section 6).

    Subject to these terms, LIST grants you a non-exclusive, non-transferable, revocable right to access and use the Application for personal, educational and non-commercial purposes. You acquire no ownership or other intellectual property rights. Without LIST’s prior written permission, you may not copy, redistribute, publicly communicate, sell or create derivative works from LIST’s own code and content.

    The name “LIST” and its logo are the property of LIST and may not be used without permission.

    6. Third-party components

    All components distributed with the software are under permissive, non-copyleft licences (MIT, ISC, 0BSD, BSD-3-Clause and Apache-2.0). They remain subject to their own licences. Copyright notices and licence texts are available in the Application under Open-source licences.

    7. Google Play

    Your download and use of the Application through Google Play are also subject to the Google Play terms of service.

    Ratings and reviews you post on Google Play are governed by Google’s terms and policies. LIST does not control how Google displays them. If you have concerns about a rating or review, you may report it through Google Play.

    8. Data protection

    LIST does not collect or process personal data through the Application, and the Application does not use advertising, analytics or tracking tools.

    If you contact LIST by email, the information you send is processed only to answer your request. Further information on how LIST processes your personal data is available in LIST’s website.

    If you download the Application from Google Play, Google processes data under its own terms and privacy notice.

    9. Disclaimer and liability

    The Application is provided “as is” and “as available”, without warranty of any kind, express or implied, including accuracy, completeness, merchantability, fitness for a particular purpose and non-infringement. LIST does not guarantee uninterrupted or error-free operation. It may suspend, modify or discontinue all or part of the Application at any time, including for maintenance, security, technical or legal reasons, and may block access in case of breach of these terms. LIST is not liable for failures or delays caused by events beyond its reasonable control.

    To the maximum extent permitted by law, LIST is not liable for any indirect, special, incidental or consequential damages, or for loss of data, arising from the use of or inability to use the Application.

    10. Applicable law and jurisdiction

    These terms are governed by Luxembourg law. The courts of Luxembourg have exclusive jurisdiction over disputes arising from these terms or the use of the Application.

    11. Contact

    Support requests and questions about the Application should be sent to LIST at jonathan.minaglou [at] list [dot] lu. LIST will respond within a reasonable time and will treat urgent issues as a priority.

    12. Changes to these terms

    LIST may amend these Terms from time to time. The current version is the version displayed in the Application. Each amended version replaces the previous one from its effective date and applies to your use of the Application from that date. Previous versions are available on request. You are invited to consult these Terms regularly, including through the Application, before using it. Your continued use of the Application after an amended version takes effect constitutes acceptance of that version, to the extent permitted by applicable law.

    Open-source licences

    Les textes juridiques sont disponibles en anglais.

    The website and the game use the following open-source components. Their authors keep their copyright, and each is used under its own licence.

    Full notices and licence texts: THIRD-PARTY-NOTICES.txt. The font licences are in OFL-Inter.txt and OFL-SpaceGrotesk.txt.

    Third-party notices (Plotly.js, canvas-confetti and others)
    SIL Open Font License – Inter
    SIL Open Font License – Space Grotesk