A new deep learning method developed by researchers at Rice University and the Baylor College of Medicine enables computers to learn about the visual world with less human instruction.
The researchers' new deep rendering mixture model largely taught itself how to distinguish handwritten digits in a dataset of 10,000 digits. The algorithm was given only 10 correct examples of each handwritten digit between 0 and 9, and then trained itself using several thousand more examples.
As a convolutional neural network, the algorithm was designed to mimic the function of biological neurons. Processing units are organized in layers, with each layer completing progressively more complex tasks.
"It's essentially a very simple visual cortex," says Rice and Baylor professor Ankit Patel. "You give it an image, and each layer processes the image a little bit more and understands it in a deeper way, and by the last layer, you've got a really deep and abstract understanding of the image."
The semi-supervised algorithm was more accurate at correctly identifying handwritten digits than almost all previous supervised algorithms, and the researchers believe artificial neural networks and semi-supervised learning algorithms can help neuroscientists better understand how the human brain works.
From Rice University
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA