Pascal Vincent

Career

Vincent has worked on artificial neural networks since 1995 and completed a PhD in computer science and machine learning at Université de Montréal under Yoshua Bengio in 2004. Vincent is a founding member of Mila – Quebec Artificial Intelligence Institute, an associate fellow in CIFAR's Learning in Machines & Brains program, and holds an appointment in Université de Montréal's Department of Computer Science and Operations Research.²¹

Vincent is a research scientist in Meta's Fundamental AI Research (FAIR) team, where the work focuses on representation learning methods that are statistically and computationally efficient and that generalize robustly out of distribution.²¹

Research

Vincent co-authored "A Neural Probabilistic Language Model" (2003) with Yoshua Bengio, Réjean Ducharme and Christian Jauvin, which learned distributed word representations jointly with a language model. Vincent's denoising autoencoders (2008 and 2010) trained networks to reconstruct inputs from corrupted versions, a pretext task that Mila's biography describes as a precursor of self-supervised learning; the 2010 JMLR paper stacked such layers to build deep networks.⁵¹³

In a December 2010 technical report, accepted for publication in Neural Computation, Vincent connected score matching with denoising autoencoders, developing the denoising score matching principle that Mila's biography notes is now used to train diffusion-based generative models. At FAIR, Vincent co-authored I-JEPA (2023), which learns image representations by predicting the representations of image regions from a context block.⁶¹⁴

Work and contributions

• I-JEPA (Image Joint-Embedding Predictive Architecture) Published research

  Co-author of "Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture" (2023), which is also listed on Vincent's Meta AI profile.⁴²

• Stacked denoising autoencoders Published research

  First author of "Stacked Denoising Autoencoders: Learning Useful Representations in a Deep Network with a Local Denoising Criterion" (JMLR, 2010).³¹

Timeline

1. 1995Began research on artificial neural networks.²
2. 2003Co-authored "A Neural Probabilistic Language Model" in the Journal of Machine Learning Research.⁵
3. 2004Completed a PhD at Université de Montréal under the direction of Yoshua Bengio.²
4. 2008Published work on denoising autoencoders with co-authors (Vincent et al. 2008).¹
5. 2010Published the JMLR paper on stacked denoising autoencoders as first author.³
6. Dec 2010Wrote the technical report connecting score matching and denoising autoencoders.⁶
7. Jan 2023Co-authored I-JEPA, self-supervised learning from images with a joint-embedding predictive architecture.⁴

Where they work

• Meta 106 people in the wiki

Related people

• Yann LeCunAlso at Meta
• Mido AssranAlso at Meta
• Yang SongAlso at Meta
• Bichen WuAlso at Meta
• Nicolas HugAlso at Meta
• Geoffrey HintonMachine-learning theory · Computer vision
• Angjoo KanazawaComputer vision
• Roy FrostigMachine-learning theory
• Cho-Jui HsiehMachine-learning theory
• Robert TibshiraniMachine-learning theory