Giving Deep Neural Networks Semantic Dementia: Using the Picture Naming Task as a Measure of Neuromorphism
Keywords
Loading...
Authors
Issue Date
2021-01-30
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Neural networks that model the way the brain works are usually evaluated based on data from exclusively
healthy patients. This project provides an additional perspective for these networks by ’lesioning’
them in correspondence with atrophy locations in the brains of Semantic Dementia patients.
Then, their behaviour on the Picture Naming task can be compared to the robust error patterns of
these patients. This allows for model comparison between two networks not based on their performance
on a test-bed of healthy brain data but in an impaired state.
For this project, two vNet models with di erent training structures were tested. While both demonstrated
a considerable degree of neuromorphism, the category-trained model performed better on
the test-bed than the model trained on a semantic objective, contrary to expectations.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen