As researchers use more and better data to train their artificial intelligence (AI) models and generate algorithms, the smarter their robots become.
Deep learning is a focus of AI research because the advantage of self-teaching systems is the removal of a great deal of manual labor. Baidu researcher Andrew Ng says deep learning is the optimal technique for ingesting and analyzing large data volumes, and much of his research at Stanford University has involved the application of machine learning to robots so they can walk, fly, and see better.
Ng says deep learning is very adept at learning features from labeled datasets, but it also is becoming proficient in unsupervised learning, in which systems learn concepts as they process large amounts of unlabeled data. Such capabilities could be helpful as we attempt to build machines that can better perceive their surroundings.
Ng says as the nexus of deep learning gravitates toward unsupervised learning, its usefulness to roboticists is likely to expand. "For a lot of applications, we're starting to run out of labeled data," he notes. For example, a robot trained to recognize 50,000 coffee mugs is a minor achievement compared to the need to improve its accuracy, which requires scaling the training datasets to millions, Ng says.
View Full Article
Abstracts Copyright © 2014 Information Inc., Bethesda, Maryland, USA
No entries found