Course group - TB2-MIAM
The Toolbox "Methods for Artificial Intelligence in Mechanics" aims at illustrating the diversity of possible applications of AI concepts by implementing them in a domain that is hardly present in the non-specialized literature. As a matter of fact, it is common to see artificial intelligence being applied to image recognition or recommendations problems, but it is more uncommon to see it applied to physical problems.
However, compressed sensing for post-processing of tomography images , Gaussian processes regression or Bayesian inference are methods that are nowadays not uncommon in research in mechanics, and it can be anticipated that they will be used by future engineers.
One of the particularities of mechanics is that this science tends to use many reliable and efficient models, based on knowledge of ancient and proven concepts. In this context, it may seem surprising to try to benefit from artificial intelligence methods, based above all on data. And actually, it seems to be nowadays commonly accepted that a method that would try to do without any physical law and to build the models only on data would be under-optimal in terms of reliability and algorithmic complexity. In other terms, knowledge-based models that have been developed in the past can be seen as an hyper-compressed and reliable form of data that should be exploited.
A recent family of approaches in computational mechanics therefore aims at using, in an optimal and complementary way, mechanical expertise and data in order to create a synergy and to solve problems that are too complex to be solved by data or expertise alone.
This course is divided in two pedagogical units that correspond to the two steps of operating an artificial intelligence :
The Toolbox has no prerequisite other than the core curriculum. It can be part of the formation of three different types on engineers :