PeopleofSuperintelligence

Researcher

Pierre-Antoine Manzagol

Pierre-Antoine Manzagol is a software developer on the Brain Team at Google Research in Montreal, and his papers run from early deep learning to machine learning for source code. In 2008 he coauthored the denoising autoencoder paper with Pascal Vincent, Hugo Larochelle and Yoshua Bengio, and in 2010 the stacked denoising autoencoders paper in JMLR. His other early work includes the TONGA natural gradient algorithm in 2007 and two studies of unsupervised pre-training in 2009 and 2010. Later he coauthored the Meta-Dataset benchmark in 2020, the PLUR framework in 2021 and a 2024 paper on resolving code review comments with machine learning. In 2025, he was among the authors of the Gemini 2.5 report.

Timeline

  1. 2007

    The TONGA algorithm

    In 2007, he coauthors the NeurIPS paper "Topmoumoute Online Natural Gradient Algorithm" with Nicolas Le Roux and Yoshua Bengio. It describes TONGA, an online approximation to natural gradient descent that converges faster than stochastic gradient descent, even on very large datasets.

  2. 2008–2010

    Denoising autoencoders

    In 2008, he coauthors "Extracting and Composing Robust Features with Denoising Autoencoders" at ICML with Pascal Vincent, Hugo Larochelle and Yoshua Bengio, which trains autoencoders to be robust to partial corruption of the input. In 2010, he coauthors the JMLR paper "Stacked Denoising Autoencoders" with Vincent, Larochelle, Isabelle Lajoie and Bengio; the stacked networks bridge the performance gap with deep belief networks and in several cases surpass them.

  3. 2009–2010

    Studies of unsupervised pre-training

    In 2009, he coauthors "The Difficulty of Training Deep Architectures and the Effect of Unsupervised Pre-Training" at AISTATS with Dumitru Erhan, Yoshua Bengio, Samy Bengio and Pascal Vincent. In 2010, he coauthors "Why Does Unsupervised Pre-training Help Deep Learning?" in JMLR with Erhan, Bengio, Aaron Courville, Vincent and Samy Bengio, whose results support a regularization explanation for the effect of pre-training.

  4. 2019–2020

    Gradient transport and Meta-Dataset

    In 2019, he coauthors a paper with Sébastien Arnold, Reza Babanezhad, Ioannis Mitliagkas and Nicolas Le Roux that proposes implicit gradient transport to reduce the variance of online optimization. In 2020, he coauthors Meta-Dataset, a large-scale benchmark of diverse datasets for few-shot classification, presented at ICLR.

  5. 2021–2024

    Machine learning for code

    In 2021, he coauthors PLUR at NeurIPS, a framework that casts 16 recently published source-code tasks as mapping a graph to a sequence of tokens and pointers. In 2024, he coauthors an ICSE-SEIP paper on resolving code review comments with machine learning, in which code-change authors at Google address 7.5% of all reviewer comments by applying an ML-suggested edit.

  6. July 2025

    The Gemini 2.5 report

    In July 2025, he is among the authors of the Gemini 2.5 report, which introduces the Gemini 2.5 Pro and Gemini 2.5 Flash models.

  7. As of October 2026

    Software developer at Google Research

    He works as a software developer on the Brain Team at Google Research in Montreal. His stated interest is machine learning methods for understanding and generating programs.

  1. Google had the most people shaping AI every year from 2014 to 2025Includes Google. Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  2. Google has the highest combined h-index among AI organizationsIncludes Google. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.
  3. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes Google. We estimate that 539 of the 6,766 people with a known current organization (8%) hold both a university role and a company role, including 20% of people at companies and 13% of people at universities.
  4. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Google. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Pierre-Antoine Manzagol yet. Send us a free-licensed photo.

  1. Topmoumoute Online Natural Gradient Algorithm | ML Anthology mlanthology.org. Checked 22 October 2026.
  2. Extracting and Composing Robust Features with Denoising Autoencoders | ML Anthology mlanthology.org. Checked 22 October 2026.
  3. Stacked Denoising Autoencoders: Learning Useful Representations in a Deep Network with a Local Denoising Criterion jmlr.org. Checked 22 October 2026.
  4. The Difficulty of Training Deep Architectures and the Effect of Unsupervised Pre-Training proceedings.mlr.press. Checked 22 October 2026.
  5. Why Does Unsupervised Pre-training Help Deep Learning? jmlr.org. Checked 22 October 2026.
  6. [1906.03532] Reducing the variance in online optimization by transporting past gradients arxiv.org. Checked 22 October 2026.
  7. [1903.03096] Meta-Dataset: A Dataset of Datasets for Learning to Learn from Few Examples arxiv.org. Checked 22 October 2026.
  8. PLUR: A Unifying, Graph-Based View of Program Learning, Understanding, and Repair proceedings.neurips.cc. Checked 22 October 2026.
  9. Pierre-Antoine Manzagol research.google. Checked 22 October 2026.
  10. [2507.06261] Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities arxiv.org. Checked 22 October 2026.