Armand Joulin

Armand Joulin is a machine-learning researcher whose work spans text classification, word representations, computer vision and foundation language models. He completed a 2012 PhD on cosegmentation at École normale supérieure and studied as an undergraduate at École Polytechnique. After a Stanford postdoctoral fellowship, he became a research scientist at Facebook AI Research. He coauthored fastText, DINO and LLaMA, and contributed to Google's privacy-focused VaultGemma model in 2025. As of October 2026, he is a Research Director at Google DeepMind responsible for Gemma, after leading Meta's EMEA work as a director.

Now at: Google DeepMind
Works: 143
Badges: Bridge

Publication numbers from OpenAlex, as of 18 October 2026. Citations and h-index are hidden because this OpenAlex record may be incomplete or mix in other authors.

Timeline

2012

A PhD in cosegmentation

In December 2012, he defends a PhD on convex optimization for cosegmentation at École normale supérieure, with Francis Bach and Jean Ponce as supervisors. He studies as an undergraduate at École Polytechnique.
2016

FastText and subword vectors

In 2016, he coauthors fastText, a text classifier that often matches deep-learning accuracy while training many orders of magnitude faster. He also coauthors a word-vector method that uses subword information and represents words absent from its training data.
March 2017

Research at Facebook AI

In March 2017, he is a research scientist at Facebook AI Research after a Stanford postdoctoral fellowship.
2021

DINO vision transformers

In 2021, he coauthors “Emerging Properties in Self-Supervised Vision Transformers.” The paper presents DINO as a self-distillation method without labels and reports 80.1% top-1 ImageNet accuracy in linear evaluation with ViT-Base.
2023

LLaMA foundation models

In 2023, he coauthors LLaMA, a collection of foundation language models ranging from 7B to 65B parameters.
2025

VaultGemma privacy model

In 2025, he contributes directly to VaultGemma, an open 1B-parameter model trained from scratch with differential privacy.
As of October 2026

Directing Gemma at Google DeepMind

As of October 2026, he is a Research Director at Google DeepMind in charge of Gemma, the open version of Gemini. Before this, he is a Research Director at Meta in charge of EMEA, where he supervises open projects such as LLaMA, DINO and fastText.