Pr Marc BUFFAT, dpt mécanique, univ. Lyon 1
%%HTML
<style>
tr,th { font-size: 22px; }
</style>
training in Modeling/Simulation
expertise in Python programming and numerical methods
acculturation and demystification of AI
public perception: (automatic translation, facial recognition)
relevance of AI in mechanics
dual expertise
Data Science / Machine Learning
Modeling in Mechanical Engineering |
Optimization problem |
|---|---|
- complex system (nddl \(\infty\)) |
- very large dimension’s problem |
- model reduction |
- non linear |
- control |
- non convex |
- sensibility |
- multi-dimensional |
- closure |
Objectives: obtain interpretable and generalizable results
based on models |
based on data |
|---|---|
explicit algorithm |
explicit algorithm |
since 1960, Finite Element computation |
today: use of Machine Learning |
in structural mechanics (linear elasticity) |
|
since 2000, High Performance Computation (HPC) |
but AI does not provide the model! |
to solve non-linear problem |
and must not be a black box |
Machine learning \(\equiv\) optimization problem
\(\rightarrow\) correlation s.t. \(\min |Y_k-F(X_k)|^2\) (but does not provide the model)
importance of the data (data base)
data cleaning / parameter selection / nondimensionalization
supervised learning for regression
no universal algorithm
Dependence of the choice of hyper-parameters
“learning by doing” approach
material constitution law (experimental database)
meteorological data-base (météo France)
data base derived from simulation (shock tube)
virtual machine (VM) : JupyterM2, JupyterGPU1
use of course management server for sharing with students and assessment
Jupyter Notebook and Jupyter Lab
available 24h/7d with a web browser (FireFox)
software environment with full feature and well-managed
AI libraries: scikit-learn, keras, tensor-flow,
database libraries: Pandas
libraries for data analysis: numpy, matplotlib, seaborn
tools: markdown, \(\LaTeX\), pandoc
It’s a set of open-source tools for interactive and exploratory computing, and a platform for creating computational narratives to engage our students in their learning process in science.
Moravec’s Paradox :
the main lesson of thirty-five years of AI research is that the hard problems are easy and the easy problems are hard
“What’s the difference bewteen AI and ML ?” (Will Wilson : nov. 2017)
-It’s AI when you’re raising money,
-it’s ML when you’re trying to hire people.
(C) France Culture 2019: l’intelligence artificielle en question