Dr Stéphane Bordas will deliver the ESE Departmental Seminar on the 2nd of February 2023: “Machine learning tools applied to mathematical models of physical phenomena: applications to medicine, sustainable energy and cultural heritage preservation”
Join us in room G41 – RSM Building – on Thursday 2nd of February 2023 at 12h10.
Or on Microsoft Teams: Stéphane Bordas Seminar
Data scarcity is a major challenge in certain fields that rely on data-driven modeling and AI, especially when system-specific data is challenging to acquire. We focus in this talk on surgical simulation, wind energy harvesting, and computational archaeology. In these fields, it is essential to harvest specific data: patient-specific data in surgical simulation, environment-specific data in wind energy harvesting, and archaeological site-specific data in computational archaeology.
To address the issue of data scarcity, real-time simulations can rely on pre-training on generic representations that can be updated as actual, system-specific data is acquired. The real-time updating of this data allows for a more accurate and patient-specific simulation.
There are numerous open problems that need to be addressed in these fields. For instance, in surgical simulation, the lack of sufficient patient-specific data hampers the development of realistic models that can better capture the complexity of surgical procedures and their outcomes. In wind energy harvesting, there is a need for more effective algorithms that can make fast predictions based on limited real-time data. In computational archaeology, the shortage of data from various sources limits the ability to uncover hidden patterns and relationships in the data.
In conclusion, data scarcity remains a major challenge in fields that rely on data-driven modeling and AI. The development of methods to acquire specific data and update generic representations in real-time is a crucial step towards advancing our understanding of complex systems and making accurate predictions. Addressing the issue of data scarcity is essential to fully realizing the potential of these technologies.
The presentation summarizes some of the key points made in the following review papers:
- From digital control to digital twins in medicine: A brief review and future perspectives
- Model selection and sensitivity analysis in the biomechanics of soft tissues: A case study on the human knee meniscus https://www.sciencedirect.com/science/article/abs/pii/S0065215622000023
- Oncology and mechanics: Landmark studies and promising clinical applicationshttps://www.sciencedirect.com/science/article/abs/pii/S0065215622000047
Surrogate machine learning-based models
- MAgNET: A Graph U-Net Architecture for Mesh-Based Simulations https://arxiv.org/abs/2211.00713
- A Graph-based probabilistic geometric deep learning framework with online enforcement of physical constraintsto predict the criticality of defects in porous materials https://arxiv.org/abs/2205.06562
- A Bayesian multiscale CNN framework to predict local stress fields in structures with microscale featureshttps://link.springer.com/article/10.1007/s00466-021-02112-3
- SOniCS: Develop intuition on biomechanical systems through interactive error controlled simulationshttps://arxiv.org/abs/2208.11676
- Probabilistic deep learning for real-time large deformation simulationshttps://www.sciencedirect.com/science/article/pii/S004578252200411X
- Quantifying discretization errors for soft tissue simulation in computer assisted surgery: A preliminary studyhttps://www.sciencedirect.com/science/article/abs/pii/S0307904X19304755
- Corotational cut finite element method for real-time surgical simulation: Application to needle insertion simulation https://www.sciencedirect.com/science/article/abs/pii/S0045782518305267
- Real-time error control for surgical simulationhttps://ieeexplore.ieee.org/abstract/document/7932498
Parameter estimation and model selection
- Digital Volume Correlation for large deformations of soft tissues: Pipeline and proof of concept for the application to breast ex vivo deformations https://www.sciencedirect.com/science/article/pii/S1751616122003952
- A tutorial on Bayesian inference to identify material parameters in solid mechanicshttps://link.springer.com/article/10.1007/s11831-018-09311-x
- Identifying elastoplastic parameters with Bayes’ theorem considering output error, input error and model uncertaintyhttps://www.sciencedirect.com/science/article/abs/pii/S0266892018300547
- Bayesian inference to identify parameters in viscoelasticityhttps://link.springer.com/article/10.1007/s11043-017-9361-0
- An open-source FEniCS-based framework for hyperelastic parameter estimation from noisy full-field data: Applicationto heterogeneous soft tissues https://www.sciencedirect.com/science/article/abs/pii/S0045794921001425
- An open source pipeline for design of experiments for hyperelastic models of the skin with applications to keloids https://www.sciencedirect.com/science/article/abs/pii/S1751616120305518
- Cortex tissue relaxation and slow to medium load rates dependency can be captured by a two-phase flow poroelastic model https://www.sciencedirect.com/science/article/abs/pii/S175161612100583X
- Digital twinning of Cellular Capsule Technology: Emerging outcomes from the perspective of porous media mechanicshttps://journals.plos.org/plosone/article?id=10.1371/journal.pone.0254512
- Accelerating Monte Carlo estimation with derivatives of high-level finite element modelshttps://www.sciencedirect.com/science/article/abs/pii/S0045782516313470
- Quantifying the uncertainty in a hyperelastic soft tissue model with stochastic parametershttps://www.sciencedirect.com/science/article/pii/S0307904X18302063
- Calculating the Malliavin derivative of some stochastic mechanics problemshttps://journals.plos.org/plosone/article?id=10.1371/journal.pone.0189994Applications to breast surgery
- Inverse deformation analysis: an experimental and numerical assessment using the FEniCS Projecthttps://link.springer.com/article/10.1007/s00366-021-01597-z
- A rigged model of the breast for preoperative surgical planning https://www.sciencedirect.com/science/article/abs/pii/S0021929021004140
About the Speaker
Stéphane is a multi-disciplinary computational and data science researcher, educator, mentor and coach. He was trained as an engineer and applied mathematician who has been teaching and researching in computational sciences since year 1999, in various capacities. He has been in the top 0.1% most cited in his field, worldwide since year 2015 (ISI Clarivate).
Stéphane leads the Legato Team (legato-team.eu), a multi-disciplinary team of about 30 researchers of a dozen nationalities. He is focusing on bringing the rigour of mathematics to bring intuition into the behaviour of complex systems. In particular, he pioneered new approaches to guarantee the quality of surgical simulation devices.
The philosophy that he has been following is to create methodologies which translate across discipline boundaries. For example, the methodological backbone of his PhD thesis supports applications in fracture mechanics, nanoscale heterogeneities, biofilm growth, cancer growth, astrocytic metabolism and many others. Recently, his team has become involved, through the Institute of Advanced Studies of the University of Luxembourg in the nascent field of Computational Archaeology.
Currently, one of the main focus points of his Team is to bring machine learning tools to bear on mathematical models of physical phenomena. In particular, his group develops adaptive data assimilation, model selection and discretisation optimisation schemes for the deformation of soft matter under large deformation with applications to surgical simulations and robotics. His team has been applying such ideas to programmable matter, multi-scale material modelling, wind energy harvesting, chemical engineering process optimisation, among others.
Stéphane has taught over 5,000 students directly and given short courses and research seminars reaching thousands of attendees. He has extensive experience in one-to-one tutoring, mentoring and coaching across various disciplines. He has directly worked with over four hundred collaborators and over fifty different companies, worldwide, as an R&D consultant. Stéphane and his students and collaborators received multiple international prizes for their research and mentorship. He has raised over 28 million euros in research funding from the private and public sector alike. He is Fellow of the Learned Society of Wales, and recipient of the 2022 Eugenio Beltrami Senior Scientist Prize. He is Editor in Chief of Advances in Applied Mechanics, Executive Editor of Data-Centric Engineering, and Subject Editor for Applied Mathematical Modelling.