Performance of basic networks trained on shape-only images.

Performance of basic networks trained on shape-only images.

<p>Accuracy of basic networks after training on 18 × 18-pixel shape-only image datasets. Error bars indicate 95% confidence intervals estimated from 10 training runs with different random initial weights. Stars indicate that accuracy is significantly above chance level (horizontal line) accord...

Full description

Saved in:
Bibliographic Details
Main Author: Christian Jarvers (6328562) (author)
Other Authors: Heiko Neumann (106738) (author)
Published: 2024
Subjects:
Science Policy
Biological Sciences not elsewhere classified
Information Systems not elsewhere classified
different learning algorithms
batch gradient descent
whereas humans base
primate visual system
local weight changes
artificial image datasets
teaching deep networks
one crucial problem
use shape features
whereas others
weight updates
use shape
primate vision
image class
training networks
make networks
widely documented
strong shape
single features
see shape
remarkably successful
network architectures
many images
large extent
feature combinations
design simple
classify images
category membership
biased towards
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:<p>Accuracy of basic networks after training on 18 × 18-pixel shape-only image datasets. Error bars indicate 95% confidence intervals estimated from 10 training runs with different random initial weights. Stars indicate that accuracy is significantly above chance level (horizontal line) according to a sign test. Circles indicate that accuracy is significantly below chance level.</p>

Similar Items