Brown University
  1. Top
  2. ...
  3. Theses and Dissertations
  4. bdr:f2hd5gs3

Towards Systematic Vision: Limits of Convolutional Neural Networks and Future Directions in Oscillatory Coding

Full Metadata

Overview

Year:
2020
Contributor:
Ricci, Matthew (creator)
Serre, Thomas (Advisor)
Bienenstock, Elie (Reader)
Frank, Michael (Reader)
Harrison, Matthew (Reader)
Brown University. Department of Cognitive, Linguistic, and Psychological Sciences (sponsor)
Genre:
theses
Subject:
dynamical systems
Cognitive science
Neural networks (Computer science)
Computational neuroscience
Deep Learning
kuramoto model
Statistical physics
Extent:
15, 250 p.

Files

Description

Abstract:
It is argued that convolutional neural networks (CNNs) struggle to represent the relations among items in a visual scene efficiently. We claim this is a deficiency in systematic reasoning and seek to ameliorate this problem with principles from neural dynamics. Theoretical and experimental evidence is provided indicating that systematic cognition depends on oscillatory cortical processes. A general machine learning procedure for phase-reduced oscillator networks is explored and applied to problems relevant to systematic reasoning.
Notes:
Thesis (Ph. D.)--Brown University, 2020

Content

Citation

Ricci, Matthew, "Towards Systematic Vision: Limits of Convolutional Neural Networks and Future Directions in Oscillatory Coding" (2020). Cognitive, Linguistic, and Psychological Sciences Theses and Dissertations. Brown Digital Repository. Brown University Library. https://repository.library.brown.edu/studio/item/bdr:f2hd5gs3/

Relations

Collection: