Ohad Dan
  • Research
  • Teaching
  • Blog
  • About
  • Research
  • Teaching
  • Blog
  • About

​​

Do deep neural networks generate visual categories like humans? The case of flower classification

6/20/2019

 
Picture
Image - Gilboa Iris (Iris haynei) passed through several deep dream generators.
I recently completed a course given by Prof. Yair Weiss ("Lab in computational neuroscience and learning") . In this course, I developed several implementations for classification algorithms of flower images using deep neural networks (DNNs). In the first phase of the project, I present an implementation that successfully classifies a novel dataset with 90% accuracy. In the second phase, I note that both humans and DNNs are excellent classifiers. Then, I use the trained classification networks of the first phase to ask whether the success humans and machines share in terms of function (both are great classifiers) also implies that the two also rely on similar mechanisms for such classification. The short answer - using the limited dataset and methods I checked - is no.
​Read the full report 
here.

    Matlab

    Archives

    June 2019
    January 2019
    February 2018
    January 2018
    February 2017
    June 2016
    May 2016
    April 2016
    March 2016

    Categories

    All
    Animation
    Art
    Computer Vision
    Graph-theory
    Image Processing
    Matlab

    RSS Feed

Powered by Create your own unique website with customizable templates.