Class Year

2019

Date

4-7-2019

Document Type

Thesis

Distinguished Thesis

Yes

Degree Name

Bachelor of Arts (BA)

Department or Program

Mathematics and Computer Science

Second Department or Program

Philosophy

First Advisor

Sugata Banergi

Second Advisor

Chad McCracken

Third Advisor

Jennifer Jhun

Fourth Advisor

Matthew R. Kelley

Abstract

The problem of fine-grained classification is one in which traditionally humans have fared better than computers. Only recently, with the advent of complex Machine Learning techniques, we have seen systems that can compete with or beat humans at this problem. In this work, we trained two Convolutional Neural Networks (CNNs) on the Stanford Dogs dataset and made them recognize dog breeds. We also analyzed the response maps of the CNNs with the aim of determining which breed-specific features the networks had learned in order to classify the images. Upon obtaining these features, we attempted to gain an insight into them for comparison with the human understanding of breeds under a Lockean interpretation.

Language

English


Share

COinS