/Optimization-strategies-for-anomaly-detection-

Project for the course "Optimization for Data Science" - Optimization strategies for anomaly detection with One class Support Vector Machines (OCSVM)

Primary LanguageJupyter Notebook

Optimization strategies for anomaly detection with One class Support Vector Machines (OCSVM)

Project for the course "Optimization for Data Science" (lecturer: Alexandre Gramfort, Master Data Science)

Aims

  • Derive the dual for the one-class SVM model,

  • implement a one-class SVM using a blackbox convex toolbox (cvxopt in Python),

  • implement your own solvers with: Proximal gradient, Coordinate Descent, Quasi-Newton,

  • present a clear benchmark of the different strategies on small and medium scale datasets.

Authors

Lea Bresson (lea.bresson@polytechnique.edu), Eya kalboussi (eya.kalboussi@polytechnique.edu)