Created in January 2021 within ICube laboratory at the University of Strasbourg, MLMS (Machine Learning, Modélisation et Simulation) team is interested in data, models and simulations for medical science and human motion. It brings together computer scientists, mathematicians, bio-mechanicians, and neuroscientists to develop functional, physical, and geometric models around a transverse axis "Assistance to medical interventions by computer". MLMS hosts the INRIA MIMESIS project-team as a subteam.
Our main research activities are:
- Developing new bio-inspired models/systems (human, organs, patients),
- Modeling and understanding of certain brain functions,
- Optimize our models and numerical simulations so that they run in real time, and
- Improve or make these models more adapted to each patient and context by exploiting real world data.
The global and transversal context around these objectives is to provide the system and assistance to physicians.
The team is structured into three themes:
- Members: Cédric Bobenrieth, Stéphane Cotin, Birgitta Dresp-Langley, Michel Duprez, Axel Hutt, Hyewon Seo
- This theme is interested in all aspects of modeling in neuroscience, geometry, biomechanics, interaction, behavior, form and movement of humans.
- We are interested in several challenges around modeling:
- Computational models of anatomy, devices and people
- Computational models of organ function (brain, heart,...)
- Interaction models
- Keywords: mathematical and geometric modeling, biomechanical modeling, interaction and behavior modeling.
Real time simulation
- Members: Stéphane Cotin, Hadrien Courtecuisse, Michel Duprez
- To implement our models, we transpose them into physical simulation form, real-time in particular.
- This is a challenging theme:
- Numerical strategies adapted to customized simulations
- Methods for real-time calculation
- Keywords: deep physics, error analysis, haptic feedback, model reduction, GPU computing
Optimization & Learning
- Members: Cédric Bobenrieth, Stéphane Cotin, Hadrien Courtecuisse, Michel Duprez, Axel Hutt, Hyewon Seo
- We aim to exploit real-world data to make our models and simulations more patient-specific and predictive.
- This theme has the following challenges:
- Extracting robust and meaningful information from signals,
- Use this information to infer the template parameters,
- Solve poorly posed problems such as shape reconstruction from partial data.
- Keywords: optimal control, shape optimization, data assimilation, deep learning.