L'éQuipe d'AppRentissage de MArseille invite les étudiants et autres personnes intéressées à ses premières portes ouvertes (POQ) le 31 mars 2017 de 12h à 15h.

One paper on ``Assessment Of Musical Noise Using Localization Of Isolated Peaks In Time-Frequency Domain'' by R. Hamon et al. has been accepted at ICASSP'17.

A PhD position (oct. 2016 — oct. 2019) on "Transfer Learning with Interacting Views" is open.

Qarma is organising the CAP conference in Marseille this year! Check the website for more information.

Imagine you are asked to answer the following questions. How can we design a computer-aided diagnosis tool for neurological disorders from multiple brain images acquired with different medical imaging devices? How can a computer identify the emotion felt by a person from his/her face and voice? How can such tools remain efficient when the data quality is poor or with missing data?
In practice, the following problem comes up: building a classifier capable of predicting the class (the diagnosis or the emotion) of a given object by taking advantage of the multiple modalities or views used to depict the objects. This is precisely what the present project aims at: the development of a well-founded machine learning framework for learning in the presence of multiple and interacting views and its confrontation to real-world problems.

This project aims at addressing these questions by gathering five partners. Picxel, a startup company leader in Europe on affective computing, and the Institut des Neurosciences de la Timone (INT, Marseille) actually face these questions in the heart of their activities. Three complementary machine learning teams from the Laboratoire d’Informatique de Paris 6 (LIP6), the Laboratoire Hubert Curien (LaHC, Saint Etienne) and the Laboratoire d’Informatique Fondamentale de Marseille (LIF, which heads the consortium), form a fundamental research network that will push the limits of the methodological state of the art.
LIVES aims at filling a hole in the current machine learning state of the art, by providing new theoretical work based on the characterization of the interactions between views, and enabling the construction of new methods and algorithms for multiview learning. This work will be carried on by a strong core of three machine learning teams, and will be confronted with real datasets provided by the two other partners. Altogether, this multidisciplinary consortium will benefit from cross-domain interactions between fundamental computer science, brain imaging, and affective computing specialists, providing new understandings for their common problems and solutions.

Journée du 15 Octobre à Saint Charles : Salles de conférences 1 et 2

Planning (prévisionnel) de la journée ici.

Ugo Louche and Liva Ralaivola,'s paper "From Cutting Planes Algorithms to Compression Schemes and Active Learning" is awarded with the best student paper prize at the 2015 International Joint Conference on Neural Networks.

We are looking for candidates in signal processing and/or machine learning for a two-year postdoc position on Audio Inpainting, funded by the French National Agency for Research (ANR), at Qarma, LIF, Marseille, France. See offer at .

The new ANR project called MAD (2015-2018) will address signal processing and machine learning problems about the inpainting of missing audio data.

We are co-organizing the LEMA workshop at ECML/PKDD'2014.

More information here.

PAutomaC: a probabilistic automata and hidden Markov models learning competition.

(Sicco Verwer, Rémi Eyraud, Colin de la Higuera) [pdf]

The 5th Asian Conference on Machine Learning (ACML2013) will be held on 13-15 November 2013, in Canberra, Australia.

  • The Multi-Task Learning View of Multimodal Data (Hachem Kadri, Stephane Ayache, Cécile Capponi, Sokol Koço, François-Xavier Dupé, Emilie Morvant)
  • On Multi-Class Classification through the Minimization of the Confusion Matrix Norm (Sokol Koço, Cécile Capponi)
  • Stability of Multi-Task Kernel Regression Algorithms (Julien Audiffren, Hachem Kadri)
  • Unconfused Ultraconservative Multiclass Algorithms (Ugo Louche, Liva Ralaivola)

Theme by Danetsoft and Danang Probo Sayekti inspired by Maksimer