Accès direct au contenu


Version anglaise


Accueil > Manifestations > Séminaires du CMLA

Séminaire du CMLA sur le thème : Optimizing to optimize

le 5 juillet 2013

par le prof. Guillermo Sapiro, Duke University (Durham, North Carolina, USA). Le prof. Sapiro est l'invité du CMLA pour quatre semaines et une série exceptionnelle de conférences. Cette première présentation sera suivie d'une autre, le 12 juillet prochain.

Site :

Parsimony, including sparsity and low rank, has been shown to successfully model data in numerous machine learning and signal processing tasks.

Traditionally, such modeling approaches rely on an iterative algorithm that minimizes an objective function with parsimony-promoting terms. The inherently sequential structure and data-dependent complexity and latency of iterative optimization constitute a major limitation in many applications requiring real-time performance or involving largescale data.

Another limitation encountered by these modeling techniques is the difficulty of their inclusion in discriminative learning scenarios. In this work, we propose to move the emphasis from the model to the pursuit algorithm, and develop a process-centric view of parsimonious modeling, in which a learned deterministic fixed-complexity pursuit process is used in lieu of iterative optimization.

We show a principled way to construct learnable pursuit process architectures for structured sparse and robust low rank models, derived from the iteration of proximal descent algorithms. These architectures learn to approximate the exact
parsimonious representation at a fraction of the complexity of the standard optimization methods.

We also show that appropriate training regimes allow to naturally extend parsimonious models to discriminative settings. State-of-the-art results are
demonstrated on several challenging problems in image and audio processing with several orders of magnitude speedup compared to the exact optimization algorithms.

The work presents a new paradigm in optimization and deep learning, and this will be briefly discussed as well.

Joint work with P. Sprechmann and A. Bronstein
Type :
Séminaires - conférences
Lieu(x) :
Campus de Cachan
Salle des Conférences, Pavillon des Jardins

Inscriptions : avant le 3 juillet ! Register before July 3rd

Prof. Guillermo Sapiro

Research Interests:

Image and video processing, computer vision, computer graphics, computational vision, biomedical imaging, brain imaging, cryo-tomography of viruses, computational tools in cryo-tomography, computational tools in early diagnosis of psychiatric disorders, differential geometry and differential equations, scientific computation, learning and high dimensional data analysis, sparse modeling and dictionary learning, applied mathematics.

Recherche d'une actualité

Recherche d'une actualité