Section: Partnerships and Cooperations

International Initiatives

Inria Associate Teams

• Hyperion: Large-scale statistical learning for visual recognition, 2012–2014

  Despite the ever-increasing number of large annotated image and video datasets, designing principled and scalable statistical learning approaches from such big computer vision datasets remains a major scientific challenge. In this associate team we collaborate with two teams of University of California Berkeley, headed respectively by Prof. Jitendra Malik and Prof. Nourredine El Karoui. It will allow the three teams to effectively combine their respective strengths in areas such as large-scale learning theory and algorithms, high-level feature design for computer vision, and high-dimensional statistical learning theory. It will result in significant progress in domains such as large-scale image classification, weakly-supervised learning for classification into attributes, and transfer learning.

Inria International Partners

• Microsoft Research NY: Zaid Harchaoui has been collaborating since the fall 2010 with Miro Dudik, formerly from Yahoo! Research (until Spring 2012), and now in the recently setup Microsoft Research New York lab, on lifted coordinate descent algorithms for large-scale learning. This collaboration lead to several published papers, including an oral presentation at CVPR 2012. Zaid Harchaoui has visited Microsoft Research NY for one week in the fall 2012. We intend to pursue this fruitful collaboration in the coming years.

• UC Berkeley: This collaboration between Bin Yu, Jack Gallant, Yuval Benjamini (UC Berkeley), Ben Willmore (Oxford University) and Julien Mairal (Inria LEAR) aims to discover the functionalities of areas of the visual cortex. We have introduced an image representation for area V4, adapting tools from computer vision to neuroscience data. The collaboration started when Julien Mairal was a post-doctoral researcher at UC Berkeley and is still ongoing, involving a student from UC Berkeley working on the extension of the current image model to videos.

• UC Berkeley, Institut Curie: In a collaboration between Jean-Philippe Vert, Elsa Bernard (Institut Curie), Laurent Jacob (UC Berkeley) and Julien Mairal (Inria LEAR) we aim to develop novel efficient optimization techniques for identification and quantification of isoforms from RNA-Seq data. Elsa Bernard was a master student between April and August 2012. She was co-advised by Jean-Philippe Vert, Laurent Jacob and Julien Mairal. Elsa Bernard has now started her PhD at Institut Curie and the collaboration is still ongoing.

• ETH Zürich: We collaborate with V. Ferrari, junior professor at ETH Zürich, and recently appointed as assistant professor at University of Edinburgh. V. Ferrari and C. Schmid co-supervised a PhD student (A. Prest) on the subject of automatic learning of objects in images and videos [3] , [9] , [20] . A. Prest was bi-localized between ETH Zürich and Inria Grenoble.

Participation In International Programs

• France-Berkeley fund: The LEAR team was awarded a grant from the France-Berkeley fund for the project with Pr. Jitendra Malik (EECS, UC Berkeley) on "Large-scale learning for image and video interpretation". The award amounts to 10,000 USD for a period of one year. The funds are meant to support scientific and scholarly exchanges and collaboration between the two teams.