Decision-making problems are ubiquitous in industry, from production optimization to in-service product operation management, including internal project and resource optimization.
In the last decades, many data-driven (e.g. Deep Reinforcement Learning) and model-based (e.g. AI Planning, Constraint Programming) approaches have been separately investigated to solve those problems.
While the former often need a huge amount of data which is scarcely available in many industrial problems, the latter are not always
suitable for problems that are complex to model with accuracy.
Also, both approaches fail to solve large problems in reasonable time and computational resources, especially in presence of uncertainty that significantly augments the combinatorial explosion of the solution space.
Principal investigators
- Sylvie Thiébaux (DR UT, LAAS-CNRS & Prof. ANU)
- Sébastien Gerchinovitz (Researcher IRT & MdC UT3, IMT)
- Romain Guillaume (MdC UT2, IRIT)
- Florent Teichteil-Königsbuch
(Researcher Airbus)
Cette chaire étudiera les techniques d’hybridation étroite entre les approches décisionnelles basées sur les données et celles basées sur les modèles en ciblant trois objectifs principaux : l’extensibilité, la robustesse et la représentativité des cas d’utilisation industriels.
It strives for opening the door to optimized, reactive and robust decision-making in
large and complex industrial problem scenarios, while significantly lowering the computation and data cost of current solvers used in the industry
The chair will gather together academic researchers from diverse institutions (LAAS, IRIT, IRT, ONERA, Ottawa University) and scientific fields, including combinatorial ptimization, search, and machine learning.
Co-chairs
-
- Christian Artigues (DR CNRS, LAAS-CNRS)
- Tommaso Cesari (Assistant Prof. Uni. Ottawa)
- Helène Fargier (DR CNRS, IRIT)
- Guillaume Povéda (researcher Airbus)
- Stephanie Roussel (CR ONERA)
Industrial partners from the aerospace and automotive industries (Airbus, Liebherr, Vitesco) will second engineers and provide challenging use cases where hybrid methods are expected to reduce operational costs due to uncertainty, model inaccuracy or solution suboptimality. This research will also benefit the health sector where we will investigate with Oncopole how to optimally schedule radiotherapy treatments for cancer patients under uncertain medical pathway appointments, with a view to improving remission chances.
- Scalability: We attack the problem of building scalable hybrid methods for sequential decision-making and combinatorial optimisation along four fronts:
- Problem-Aware Deep Learning Architectures: we first design more frugal deep learning architectures that exploit the problem structure to learn (partial) solutions and heuristics to guide search.
- Properties of Architectures and Learnt Models we then provide guarantees about the representative power,
approximation and generalization capabilities, and about other properties of the models they learn - Effective Data Generation: we follow by developing effective training strategies that work in synergy with search.
- Guiding Search: : élaboration des méthodes de recherche et d’optimisation qui utilisent efficacement les conseils appris pour nos cas d’utilisation, et des outils qui nous permettent d’explorer l’espace plus large des méthodes hybrides.
- Robustness:
- Learning-based decomposition methods we will learn a good selection of scenarios so as to minimize the number of times the master problem and the adversarial subproblems are solved.
-
Hybrid methods for proactive-reactive optimisation how and when to rebuild a robust solution (i.e. how and when to react), when unexpected disruptions push the current solution out of the predicted scenario set.
Scientific outcomes
scalable, data-frugal, and robust hybrid methods for search and combinatorial optimisation
new methods for injecting constraints and domain knowledge in learning models
theoretical and empirical results on generalization/approximation properties of deep learning models
Industrial outcomes
Manufacturing: decision making tools that take into account uncertainty and more complex constraints than is possible today, resulting in manual replanning decreasing by 50%, storage costs of plants reducing by 20%, and the adherence to the schedule doubling
Space: deployment of an algorithm with improved image acquisition ordering and weather knowledge injection, resulting in a 5% increase in high-quality image production
Health: decision making tool integrated in Oncopole planning tools, reducing by 80% the need of reallocating machines to patients, and by 50% the planning work of people in charge.