IC Colloquium : Similarity search meets GPUs and learning
Event details
| Date | 27.11.2017 |
| Hour | 16:15 › 17:30 |
| Location | |
| Category | Conferences - Seminars |
By : Hervé Jegou - Facebook AI Research
Video of the talk
Abstract :
This talk will discuss how high efficient nearest neighbor search can leverage several learning computer vision and natural language processing tasks. After briefly presenting some works conducted at FAIR on similarity search on CPUs and GPUs, I will present recent work on semi-supervised learning at a massive scale. Then I will present a few works related to natural language processing involving similarity search strategies. In particular, I will present a work employing nearest neighbor matching and adversarial training to perform word translation without any parallel data.
Bio :
Herve Jégou is a Research Manager with Facebook AI Research since 2015. He is a former student of the Ecole Normale Supérieure de Cachan, holding a M.S. (2002) and PhD (2005) from University of Rennes I. During his PhD, he worked on source/channel coding. After that, he turned out to Computer Vision and Pattern Recognition and joined INRIA as a permanent researcher in 2006, mostly working and leading several projects related to large image and video collections. His current interests at FAIR include large-scale processing, computer vision and natural language processing.
More information
Video of the talk
Abstract :
This talk will discuss how high efficient nearest neighbor search can leverage several learning computer vision and natural language processing tasks. After briefly presenting some works conducted at FAIR on similarity search on CPUs and GPUs, I will present recent work on semi-supervised learning at a massive scale. Then I will present a few works related to natural language processing involving similarity search strategies. In particular, I will present a work employing nearest neighbor matching and adversarial training to perform word translation without any parallel data.
Bio :
Herve Jégou is a Research Manager with Facebook AI Research since 2015. He is a former student of the Ecole Normale Supérieure de Cachan, holding a M.S. (2002) and PhD (2005) from University of Rennes I. During his PhD, he worked on source/channel coding. After that, he turned out to Computer Vision and Pattern Recognition and joined INRIA as a permanent researcher in 2006, mostly working and leading several projects related to large image and video collections. His current interests at FAIR include large-scale processing, computer vision and natural language processing.
More information
Practical information
- General public
- Free
- This event is internal
Contact
- Host : Pascal Fua