Advanced Machine Learning


Teacher coordinatorStéphane Canu
Teacher(s)Stéphane Canu, Alain Rakotomamonjy
TeachingLectures : 18h Exercises : 18h
OptionData science
Web site

Aim and objective

This course aims at providing advanced notions in machine learning related to dictionary learning for signal and image representation, matrix factorization for recommendation system and adapted recent convex optimization methods. adapted to this task

Outcome learning

  • INSA reference data :
    • Concevoir un système d'ingénierie des données [3P]
    • Diagnostiquer des erreurs dans des données [3P]
  • CNISF reference data :
    • J10P [2P]
    • J40K [1P]
1 - Notion, 2 - Concept, 3 - Application, I - fully, P - incomplete

Course description

  • Introduction
    • Audio source separation
  • Regularization techniques
    • L2 vs L1 Regularisation
    • Illustration example of L1 regularization or penalty
    • Notions of subgradient and Fenchel duality
  • Lagrangian duality
  • Sparse regression
    • Ridge regression
    • Lasso problem
  • Introduction to proximal methods
  • Dictionary learning and matrix factorization
    • Alternate methods of optimization
    • Alternate ISTA algorithm as a particular example
  • Explored application domains during the course
    • Image restoration
    • Recommendation systems
    • Signal denoising and inpainting applications


Data Mining, Statistics, Signal Processing


  • "Convex optimization" S. Boyd and L. Vandenberghe
  • "Proximal algorithms" N. Parikh and S. Boyd dans Foundations and Trends in Optimization, 1(3):123-231, 2013


  • Final exam: 50%
  • Project: 50%