Interdisciplinary Institute in Artificial Intelligence

DESCRIPTION

Motivation and expected impact: The goal of the chair is to develop a new interactive learning framework that incorporates expert feedback into the training phase for exploring the multidimensional design space of materials. Designing new materials is at the heart of any technological transition, and many such transitions lie ahead: lighter transportation (more efficient structures), energy production (materials for renewables), the circular economy (recyclable materials), the resource crisis (substitution of critical chemical species), and gas capture and release (CO2, toxic gases). The immense scope of the materials design space presents a key opportunity to combine high-throughput materials exploration—using experimental and simulation strategies—with advanced artificial intelligence tools. We anticipate new breakthroughs in designing materials with improved compositions and properties that address the needs of future society and environmental challenges.

Scientific objectives and context: Designing new materials with optimized functions is essential to address the industrial challenges posed by future societal constraints. Targeted functions include structural, chemical, and physical properties, as well as those related to safety, environmental sustainability, recycling, low cost, and accessibility. The search for new materials with specific properties remains largely empirical, traditionally guided by intuitive and trial-and-error methods. To move beyond this, data-to-knowledge strategies are showing tremendous promise in materials science. The aim of the proposed research is to establish new machine learning (ML) approaches in order to develop real materials based on specific properties (local structure, microstructure, thermodynamic, and mechanical properties), to predict new phases, and to determine the structure of materials at the atomic scale.

ACTIVITIES

The program chair has been organized into research areas, each involving collaborative work between computer science (LIG), materials science (SIMAP), and mathematics (IF) teams, with additional links to the proposed "Towards More Data Efficiency in Machine Learning" chair within the MIAI institute.

The chair's activities include the development of new approaches and frameworks through two Ph.D. projects conducted in collaboration with the SIMAP, LIG, and IF laboratories, as well as applied research projects in collaboration with the chair's industrial partners: 3 master's theses, 2 Ph.D. projects in progress, 1 Ph.D. project planned for fall 2021, and 1 postdoctoral position planned for early 2021.

The chair's industrial partners to date include Atos, Mentor, Constellium, and ArcelorMittal. Discussions are underway to explore new collaborations.

CHAIR EVENTS

Organization of the 17th International Conference on Aluminum Alloys (ICAA17) in Grenoble, October 2020—converted to a virtual event (A. Deschamps, chair of the organizing committee)

RESEARCH TOPICS

• Development of new transfer learning approaches to solve new machine learning problems more efficiently by leveraging knowledge gained from solving other related problems, using latent variables that represent unobserved physical behaviors – 2 PhDs in collaboration with Mentor and Atos.
• Description and classification of the material’s atomic structure. Advanced machine learning approaches will be used to validate low-computational-cost interatomic potentials using ab initio calculation databases. Understanding local many-body 3-D configurations requires advanced neural network models trained on partially labeled datasets obtained through molecular dynamics—1 Ph.D. in collaboration between SIMAP, LIG, and IF.
• The materials design space comprises 10 or more dimensions of chemical composition and processing parameters, resulting in complex atomic structures such as metal-organic frameworks (MOFs) or complex mesoscale microstructures in metallic alloys. New machine learning approaches will be developed to guide descriptive models and property optimization based on combinatorial experimental datasets covering discrete parts of the design space—1 PhD position to be filled with Constellium and 1 postdoctoral position to be filled with ArcelorMittal.
• New machine learning tools based on topological data analysis will be developed to actively improve the selection of informative labeled observations in conjunction with semi-supervised algorithms, with a view to examining the topological atomic-scale structure of a broad class of materials—1 PhD position in collaboration between SIMAP and LIG.

More about this chair : Home - MAGNET (liglab.fr)