Impressive but highly technical research in the field of image recognition carried out at the secretive facility known as Google X and released to the public Tuesday may not yet have a commercial application to the company, but it fits into Google‘s intense interest in machine learning and connections between items of data – which has already been released to the public in the form of the Knowledge Graph, an augmented search tool.
“You probably use machine learning technology dozens of times a day without knowing it – it’s a way of training computers on real-world data, and it enables high-quality speech recognition, practical computer vision, email spam blocking and even self-driving cars,” wrote Google Fellow Jeff Dean and Stanford researcher Andrew Ng. “But it’s far from perfect – you’ve probably chuckled at poorly transcribed text, a bad translation or a misidentified image.”
In this round of Google X experiments, researchers built a neural algorithm that learns more like a newborn baby than previous attempts (they also used Google’s data centers to build an extremely large, powerful neural network). Instead of feeding it information already placed into categories, they let it watch unlabeled YouTube videos for a week – and what it learned was surprising, though perhaps predictable given the culture of the web.
“Remember that this network had never been told what a cat was, nor was it given even a single image labeled as a cat,” they wrote. “Instead, it ‘discovered’ what a cat looked like by itself from only unlabeled YouTube stills. That’s what we mean by self-taught learning.”
A tangentially related product already available to Google users is Knowledge Graph, a semantic network that calls up information in the form of connections between topics culled from the web and other sources.
It’s also another example of increasingly heavyweight research by Google in diverse fields. The team that developed the technology will present a paper based on the research during the International Conference on Machine Learning at the end of the month.
However, it fits tightly into the company’s vision for categorizing knowledge in the future.
“We believe machine learning could be far more accurate, and that smarter computers could make everyday tasks much easier,” Dean and Ng wrote.