thumbnail image

  • Home
  • About
  • Research
  • Awards
  • Press
  • Talks
  • Home
  • About
  • Research
  • Awards
  • Press
  • Talks
  • Powered By
    Strikingly
    • Yann N. Dauphin

      Research Scientist

    • About

      Yann N. Dauphin is a machine learning researcher in the Google Brain team working on understanding the fundamentals of deep learning algorithms and leveraging that in various applications. He has published seminal work on understanding the loss surface of neural nets. Prior to joining Google in 2019, he was a researcher at Facebook AI Research from 2015 to 2018 where his work led to award-winning scientific publications and helped improve automatic translation on Facebook.com. He received his PhD in 2015 from U. of Montreal under the supervision of Prof. Yoshua Bengio. During this time, he and his team won international machine learning competitions such as the Unsupervised Transfer Learning Challenge in 2013.

       

      Email: yann@dauphin.io Resume: Link

    • Research

      Here are selected publications. Full list available on Google Scholar.

      Identifying and attacking the saddle point problem in high-dimensional non-convex optimization

      Published in Neural Information Processing Systems 2014

      This paper dispels the myth of bad local minima in high dimension and shows that the loss surfaces of neural networks have remarkable properties. This seminal paper has helped renew interest for understanding non-convex optimization as well as the study of neural nets using methods from statistical physics.

      mixup: Beyond empirical risk minimization

      Published in International Conference on Representation Learning 2017

      Large deep neural networks are powerful, but exhibit undesirable behaviors such as memorization and sensitivity to adversarial examples. This work proposes a remarkably simple learning principle called mixup to alleviate these issues by encouraging the network to be linear between examples.

      Convolutional sequence to sequence learning

      Published in International Conference on Machine Learning​ 2017

      The prevalent approach to sequence to sequence problems such as translation was recurrent neural networks. In this work, we introduced a state-of-the-art architecture based entirely on convolutional neural networks. Compared to recurrent models, computations over all elements can be fully parallelized during training to better exploit the GPU hardware.

    • Awards

      Here is a selected list of awards I've received.

      Best paper award

      IEEE SPS 2020

      Best paper Honorable mention

      NeurIPS 2011

      First place

      Unsupervised Transfer Learning Challenge (Phase 2)

      Best paper Honorable mention

      ACL 2018

      First place

      Emotion Recognition in the Wild Challenge​ 2013

    • Press

      Facebook's new AI could lead to translations that actually make sense

      Wired

    • Talks

      Tangent Propagation

      At the International Conference on Machine Learning (ICML).

      Manifold Tangent Classifier

      At the Neural Information Processing Systems (NIPS).

    © 2019

    Create a site with
    This website is built with Strikingly.
    Create yours today!

    This website is built with Strikingly.

    Create your FREE website today!

    Cookie Use
    We use cookies to ensure a smooth browsing experience. By continuing we assume you accept the use of cookies.
    Learn More