The art of modeling in mechanics of solid materials, quite often guided by the general principles of thermodynamics, is mainly a craft where the researcher or the engineer is seeking a good description (variables and functions) of a partially observed phenomenon. This active field of research has been recently renewed with the help of machine learning tools, especially neural networks and sparse or symbolic regression.

This session aims to overview the topic of automatic learning of constitutive relations in solid mechanics. Applications, as well as more fundamental research, are welcome. In the literature, two main applications are currently developed: synthesizing macroscopic constitutive relations of numerical representative volume elements, and modeling macroscopic stress-strain curves; this mini-symposium includes, but is not limited to these applications. Note that, while synthetic data are most often used, the evaluation of strategies based on real data obtained from various experimental techniques is highly welcome. The experimental/model/simulation triptych being a whole, the interactions between the modeling itself and experimental or simulation constraints are of interest in this session. Topics covered within this session include but are not limited to:

- Data: synthetic/real data, frugality, informational content…
- Learning strategies: supervised, unsupervised, reinforcement learning…
- Physics framework and integration: CANN, GENERIC, PINN, PANN, TANN…
- AI: dedicated loss functions, structure, initialization, hyper-parameter tuning, explicability, robustness…
- Sparse regression, Bayesian learning, symbolic regression
- Interpretability and fulfilment of physical constraints
- Acceleration of multiscale and multiphysics simulations using machine learning techniques