Learning Sparse Penalties for Change-Point Detection using Max Margin Interval Regression

Guillem Rigaill (1) , Toby Dylan Hocking (2, 3, 4) , Francis Bach (2, 5) , Jean-Philippe Vert (4, 3)

1 MIA-Paris - Mathématiques et Informatique Appliquées
2 SIERRA - Statistical Machine Learning and Parsimony
3 CBIO - Centre de Bioinformatique
4 Cancer et génome: Bioinformatique, biostatistiques et épidémiologie d'un système complexe
5 LIENS - Laboratoire d'informatique de l'école normale supérieure

In segmentation models, the number of change-points is typically chosen using a pe- nalized cost function. In this work, we pro- pose to learn the penalty and its constants in databases of signals with weak change-point annotations. We propose a convex relaxation for the resulting interval regression problem, and solve it using accelerated proximal gra- dient methods. We show that this method achieves state-of-the-art change-point detec- tion in a database of annotated DNA copy number profiles from neuroblastoma tumors.

Domaines: Apprentissage [cs.LG]; Bio-informatique [q-bio.QM]; Bio-Informatique, Biologie Systémique [q-bio.QM]

Guillem Rigaill, Toby Dylan Hocking, Francis Bach, Jean-Philippe Vert. Learning Sparse Penalties for Change-Point Detection using Max Margin Interval Regression. ICML 2013 - 30 th International Conference on Machine Learning, Supported by the International Machine Learning Society (IMLS), Jun 2013, Atlanta, United States. ⟨hal-00824075⟩