Analysing Audio and multimodal data with Nonnegative Matrix Factorisation
Prof. Slim Essid
Telecom ParisTech, Paris
DEIB - "Beta" Room, Building 24 (Via Golgi, 40 - Milano)
February 22nd, 2017
10.00 am
Contacts:
Augusto Sarti
Research Line:
Signal processing for multimedia and telecommunications
Telecom ParisTech, Paris
DEIB - "Beta" Room, Building 24 (Via Golgi, 40 - Milano)
February 22nd, 2017
10.00 am
Contacts:
Augusto Sarti
Research Line:
Signal processing for multimedia and telecommunications
Sommario
Slim Essid will present an overview of research performed at the Audio group of Telecom ParisTech, focusing on contributions to Nonnegative Matrix Factorisation (NMF) and applications to audio and EEG data analysis.
First he will briefly cover our recent work on feature learning with NMF where we have explored diverse algorithmic variations that allow us to achieve competitive classification results in audio scene analysis tasks.
Then prof. Slim Essid will describe co-factorisation schemes for the analysis of temporally structured multimodal data (including videos), where his research group has introduced original formulations and algorithms to perform smooth factorisation and soft co-factorisation. The former yields representations suitable for the analysis of temporal sequences, possibly with piecewise-constant activations, useful in many temporal segmentation tasks. As for the soft co-factorisation schemes, they allow for jointly performing two (or more) factorisations, so as to produce models capturing the dependencies that may exist between the modalities being analysed in parallel.
First he will briefly cover our recent work on feature learning with NMF where we have explored diverse algorithmic variations that allow us to achieve competitive classification results in audio scene analysis tasks.
Then prof. Slim Essid will describe co-factorisation schemes for the analysis of temporally structured multimodal data (including videos), where his research group has introduced original formulations and algorithms to perform smooth factorisation and soft co-factorisation. The former yields representations suitable for the analysis of temporal sequences, possibly with piecewise-constant activations, useful in many temporal segmentation tasks. As for the soft co-factorisation schemes, they allow for jointly performing two (or more) factorisations, so as to produce models capturing the dependencies that may exist between the modalities being analysed in parallel.
Biografia
Slim Essid is a Full Professor at the Department of Image and Signal Processing-TSI of TELECOM ParisTech and the head of the AAO group. He received the state engineering degree from the École Nationale d’Ingénieurs de Tunis in 2001; the M.Sc. (D.E.A.) degree in digital communication systems from the École Nationale Supérieure des Télécommunications, Paris, France, in 2002; the Ph.D. degree from the Université Pierre et Marie Curie (UPMC), in 2005; and the habilitation (HDR) degree from UPMC in 2015.
He has been involved in various French and European research projects among which are Quaero, Networks of Excellence Kspace and 3DLife, and collaborative projects REVERIE and LASIE.
He has been involved in various French and European research projects among which are Quaero, Networks of Excellence Kspace and 3DLife, and collaborative projects REVERIE and LASIE.