Matthijs Douze
Research scientist at Meta FAIR; implemented most of the CPU side of the Faiss similarity search library
Matthijs Douze is a research scientist at Meta's Fundamental AI Research (FAIR) lab, where Douze has worked since November 2015 on large-scale indexing, machine learning with graphs, and similarity search for images and videos. The Faiss README lists Douze among the library's main authors as the person who implemented most of the CPU version, and Douze is first author of the 2024 paper "The Faiss library". From 2005 to 2015 Douze worked in the LEAR team at INRIA on image and vector indexing and video search.135
Career
From 2005 to 2015 Douze worked in the LEAR team at INRIA on image indexing, large-scale vector indexing, event recognition in videos and similar-video search. Between 2010 and 2015 Douze also managed Kinovis, a 3D motion capture studio at INRIA, and developed geometric algorithms for constructive solid geometry.1
Douze joined Facebook AI Research (FAIR) as a research scientist in November 2015, working on large-scale indexing, machine learning with graphs, and similarity search on images and videos.1
Faiss and research
The Faiss repository, created in February 2017 under the MIT license, credits Hervé Jégou with initiating the project, Douze with implementing most of the CPU Faiss, and Jeff Johnson with the GPU Faiss. Douze co-authored "Billion-scale similarity search with GPUs" (2017) and was first author of "The Faiss library" (2024).345
Douze's other FAIR publications include work on deep clustering for unsupervised visual features (2018), the Stable Signature method for watermarking latent diffusion models (2023), and research showing that watermarking can make language models "radioactive" (2024).1
Work
FaissOpen-source code
Listed in the Faiss README as having implemented most of the CPU Faiss, and first author of "The Faiss library" (2024).35
Billion-scale similarity search with GPUs (external site: arxiv.org)Published research
Co-author, with Jeff Johnson and Hervé Jégou, of the 2017 paper on GPU similarity search that the Faiss README asks users to cite for GPU Faiss.43
Career timeline
2005
Joined the LEAR team at INRIA, working on image and large-scale vector indexing.1
Nov 2015
Joined Facebook AI Research (FAIR) as a research scientist.1
Feb 2017
Co-authored "Billion-scale similarity search with GPUs" as the Faiss repository was opened.43
Sep 2018
Co-authored "Deep Clustering for Unsupervised Learning of Visual Features".1
Sep 2023
Co-authored "The Stable Signature", on watermarking latent diffusion models.1
Jan 2024
Posted "The Faiss library" as first author.5
Other documented roles
Roles the sources document but that are not confirmed as current.
- Researcher in the LEAR team (2005–2015); manager of the Kinovis motion capture studio (2010–2015), INRIA1
Related people
More sourced profiles at the same organization, in the same field, or from the same school.
- Adina WilliamsAlso at Meta
- Alexander ToshevComputer vision
- Alban DesmaisonAlso at Meta
- Fei-Fei LiComputer vision
- Alexandr WangAlso at Meta
- Matt DeitkeAlso at Meta
- Alexandre GramfortAlso at Meta
- Matthew ZeilerComputer vision
- Andrew BosworthAlso at Meta
- Piotr DollárAlso at Meta
- Christoph FeichtenhoferAlso at Meta Superintelligence Labs
- Tsung-Yi LinComputer vision
Is this you?
Profiles cover public professional work only, and every line cites a source. If something is out of date, missing or wrong, send a correction with a link to an official page and we will update it. Suggest a correction
Add this badge to your website or GitHub profile README. It links back to this page. It is not an endorsement or a rating.
Markdown (for a README)
[](https://unitedstatesofamericasuperintelligence.com/people/matthijs-douze/)
HTML
<a href="https://unitedstatesofamericasuperintelligence.com/people/matthijs-douze/"><img src="https://unitedstatesofamericasuperintelligence.com/badges/profile.svg" alt="Matthijs Douze — Profile on USASI" height="20"></a>
This profile covers public professional work only, from the sources below. It is not an endorsement or a ranking. To correct or remove it, use the correction form.