Florent Perronnin

  • machine learning
  • computer vision
  • artificial intelligence
  • management

I am Director of Science and Engineering at NAVER LABS Europe in Grenoble, France.
My role involves the scientific coordination of our research activities around Artificial Intelligence (AI) including machine learning, computer vision and language technologies.

I joined the lab in this role in July 2016. From May 2015 to June 2016, I was the Director of the Facebook AI Research (FAIR) Lab in Paris. From 2005 to 2015, I was with the Computer Vision group of Xerox Research in Grenoble, France, first as a Research Scientist then as Principal Scientist and Manager. From 2000 to 2001, I was a Research Engineer with Panasonic in Santa Barbara, CA, working on speech and speaker recognition. I received my Engineering degree from Telecom ParisTech in 2000 and my PhD in Computer Science from EPFL in 2004.

My research has found applications in several business verticals including Business Process Outsourcing (BPO), Transportation, Retail or Communication and Marketing. It also won several awards in public competitions such as PASCAL VOC, ImageNet and ImageCLef.

The research topics I am currently most passionate about include representation learning, lifelong learning and machine reasoning.


  • INRIA and NAVER LABS Europe are co-organising PAISS, an Artificial Intelligence Summer School,  July 2-6 in Grenoble, France. 
  • The Xerox Research Centre Europe (XRCE) became Naver Labs Europe (NLE)!
  • Together with Adrien Gaidon (TRI) and Antonio López (CVC), we are co-organizing a special IJCV issue on Synthetic Visual Data for Computer Vision.
  • I will be industry co-chair for ICCV 2017 in Venice, Italy.
  • New TPAMI (with Naila Murray, Hervé Jégou and Andrew Zisserman) on "Interferences in match kernels". Paper on ieeexplore.
  • I gave an invited tutorial on "Output embedding for large-scale computer vision" at ECCV 2016: (slides 6.66MB)  
  • 1 paper at ECCV (with Matthijs Douze and Hervé Jégou) on very large-scale approximate search with "Polysemous codes". Paper on arXiv