Teaching Machine Learning
For the second year, I am teaching an introduction course to Machine Learning from mechanical engineers.
The point is to gave an overview on the field in less than eight hours.
I made a 1,5h lecture, 2h exercise and a 4h practical works.
The lecture course cannot detail many things. I choosed to emphasize on the classical Support Vector Machine but powerful, with which students can use their linear algebra skills, and on neural networks methods to introduce Deep Learning.
Then in the supervised exercise, students practices with both methods.
Finally, in the four hours practical works, they work on a more complete pipeline to solve a concrete problem.
Last year, in addition to this framework, I tried to overview differents problems in machine learning: supervised learning, unsupervised learning/dimensionnal reduction, reinforcement learning.
It was too much for the mean student.
Only two out of thirty hung on all subjects.
The practical work was on reinforcement learning using OpenAI Gym.
This was really fun to prepare.
This year, I choose to only introduce supervised learning. I also change the practical works subjects to hook everyone. Students will do something not so ethical: human faces beauty regression, using the strange SCUT-FBP dataset.
It was a great fun to design student proofed Jupyter Notebooks and Google Colab, while doing my slides with Marpit.
Teaching is a research topics on his own.
_The course material is in my MECA653 ML4Mecha repository.
The Polytech Annecy-Chambéry MECA653 module repository is here.
