September 2019
Volume 19, Issue 10
Open Access
Vision Sciences Society Annual Meeting Abstract  |   September 2019
Examining Class Dependant Sub-Paths in Deep Neural Networks
Author Affiliations & Notes
  • Mathew Monfort
    CSAIL, MIT
  • Kandan Ramakrishnan
    CSAIL, MIT
  • Alex Andonian
    CSAIL, MIT
  • Aude Oliva
    CSAIL, MIT
Journal of Vision September 2019, Vol.19, 28b. doi:https://doi.org/10.1167/19.10.28b
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Mathew Monfort, Kandan Ramakrishnan, Alex Andonian, Aude Oliva; Examining Class Dependant Sub-Paths in Deep Neural Networks. Journal of Vision 2019;19(10):28b. doi: https://doi.org/10.1167/19.10.28b.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Understanding the decision of an artificial neural network is an important task when we consider how these models are being applied to problems such as robotics, autonomous driving and other increasingly important areas where an incorrect inference can lead to costly outcomes. Recent work in the area disentangles the learned representations in order to identify key concepts that are being learned and quantifies the dependence of classes on specific feature directions. Most of these approaches focus on what the models are learning. We propose a method to understand why the models are learning these concepts and feature dependencies. We extend the ideas of feature selectivity and network dissection to identify learned class relationships and build quantifiable feature-class associations based on learned network parameters. This approach allows us to not just understand what a model is learning but to also gain insight into why it is learning these concepts and how it is using them. We apply these methods to multiple network architectures trained for different tasks and explore ways to regularize feature dependency to improve generalization.

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×