Communications of the ACM
Dataset Bridges Human Vision and Machine Learning
A new dataset allows cognitive neuroscientists to better leverage the deep learning models that have significantly improved artificial vision systems.
Credit: alltechasia.com
Researchers at Carnegie Mellon University and Fordham University have created a new dataset that allows cognitive neuroscientists to better leverage the deep learning models that have significantly improved artificial vision systems.
The dataset, called BOLD5000, is made up of brain scans of four volunteers who each viewed 5,000 images taken from two other computer vision datasets: ImageNet and COCO.
While BOLD5000 is much larger than other neuroimaging studies, which often have utilize 100 or fewer images, some researchers think it is still too small.
However, the field's response has been positive so far, as the publicly available BOLD5000 dataset has already been downloaded more than 2,500 times.
A reasonable fMRI dataset would require at least 50,000 stimulus images and many more volunteers to make any progress, since the class of deep neural networks used to analyze visual imagery is trained on millions of images.
From Carnegie Mellon University
View Full Article
Abstracts Copyright © 2019 SmithBucklin, Washington, DC, USA
No entries found