Within machine learning, neural networks are all the rage. But Google researcher Geoff Hinton recently published two research papers that could change that. Hinton’s papers showed how he used capsule networks — an idea that he has been working on since the 1970s — to create a system that was better at some types of computer vision than today’s neural networks are.
Human brains have areas that are specifically designed to be good at things like vision or speech. Hinton’s capsule network mimics that design by creating a system that is pre-wired to be good at computer vision.
Some other experts in AI believe this new design could be revolutionary.