September 26-27, 2016
New York, NY

Deep learning: Modular in theory, inflexible in practice

3:45pm–4:25pm Monday, 09/26/2016
Implementing AI
Location: 3D10 Level: Advanced
Average rating: *****
(5.00, 3 ratings)

Prerequisite knowledge

  • A high-level understanding of sophisticated architectures
  • A background in software engineering (especially graph algorithms)
  • What you'll learn

  • Understand the challenges that limit the capability of today's research in deep learning
  • Explore a practical attempt at a deep learning library of the future
  • Description

    The high-level view of deep learning is elegant: composing differentiable components together trained in an end-to-end fashion. The reality isn’t that simple, and the commonly used tools greatly limit what we are capable of doing. Diogo Almeida explains what we can do about it and offers a practical attempt at a deep learning library of the future.

    Topics include:

    • An overview of the large body of existing research showing the diversity of models that have been created with deep learning
    • Theoretical modularity and how deep learning could solve many hard AI problems, including ones we haven’t even attempted yet
    • The problems with existing libraries and how these challenges limit the capability of today’s research
    • Examples of the kinds of architectures one would want, which are difficult to implement with today’s tools
    • The requirements of a “deep learning library of the future”
    • Specific details about a real-world library that follows that template
    • Practical examples of the things we have been able to do with that library
    Photo of Diogo Moitinho de Almeida

    Diogo Moitinho de Almeida

    Enlitic

    Diogo Moitinho de Almeida is a data scientist, software engineer, and hacker. Currently, Diogo is a senior data scientist at Enlitic, where he works to radically improve the quality of medical diagnosis using deep learning, advance the state of the art in modeling, and build novel ways to interact with neural networks. Previously, he was a medalist at the International Math Olympiad, ending a 13-year losing streak for the Philippines; received the top prize in the Interdisciplinary Contest in Modeling, achieving the highest distinction of any team from the Western Hemisphere; and won a Kaggle competition, setting a new state of the art for black box identification of causality and getting the opportunity to speak at the Conference on Neural Information Processing Systems.

    Comments on this page are now closed.

    Comments

    Ebraheem Fontaine
    09/27/2016 6:42am EDT

    Can you please share your slides with the comprehensive overview of references to recent advances?