Supervision: Stephane Marchand-Maillet

Date of proposal: Nov. 2023

In a continuing effort to create interactive pedagogic tools to teach Machine Learning, this project aims at creating an interactive (web) demo of a theorem/principle/algorithm in Machine Learning. Topics include:

• Optimization (eg Gradient descent, backprop, SGD)
• Linear Algebra (eg eigen vectors, SVD)
• Probabilities (eg sampling, change of variables)
• Function analysis (eg Fourier, wavelets)
• Concentration (eg distances, angles, WLLN, CLT)
• Classical algorithms (eg ID3, Minimax, LogReg, Naive Bayes)
• Visualization (eg NN, SVMs, clustering)
• Manifold learning (eg tSNE, DiffMaps, UMAP, SOM)
• Learning theory (eg Universal approximation, PAC learning, Hoeffding inequality)
• …

The aim is to work out the topic “to the bone” to create a fundamental understanding and a high-level (intuitive) picture, then instantiated as an interactive demo (graphical or not). See eg decision trees (thanks to R. Maggio-Aprile and E. Icet, 2020 - A. Erne, 2022)

There is therefore material to accommodate several students. Each project will be adapted to the level (App Info, Bachelor, Master,..)

If interested please contact me.