This item is under embargo and not available online per the author's request. For access information, please visit http://libanswers.wustl.edu/faq/5640.

Date of Award

Spring 5-4-2021

Author's School

McKelvey School of Engineering

Author's Department

Computer Science & Engineering

Degree Name

Master of Science (MS)

Degree Type

Thesis

Abstract

Do artificial neurons in CNNs learn to represent the same visual information as the biological neurons in primate brains? Previous studies have shown that the visual recognition pathway (ventral stream) in humans and monkeys increasingly represents animate objects [16]. We used a heatmap attribution technique borrowed from convolutional neural networks to generate biological feature maps identifying regions in scenes that elicit responses from neurons along the ventral stream (V1/V2, V4, and IT). Biological feature maps were then compared to activation maps produced by units in convolutional neural networks. We found that image regions containing animals elicited increasingly larger responses along the ventral stream, while such animacy features are not represented in artificial neural networks.

Language

English (en)

Chair

Roch Guérin

Committee Members

Dr. Chien-Ju Ho Dr. Ulugbek Kamilov Dr. Carlos Ponce

Available for download on Tuesday, October 26, 2021

Included in

Engineering Commons

Share

COinS