What Is Capsule Network (CapsNet)?

Capsule networks are like the cool kid in school who everyone knows but no one really knows. They're all about image processing, and they've been around for a while. What makes them so special? Let's start with the fact that they were developed by Geoffrey Hinton, who pioneered deep learning algorithms. So you know this is some serious stuff. Then there's their methodology: capsule networks try to understand objects from a three-dimensional spectrum by applying specific techniques that allow them to focus on particular aspects of an image and identify patterns within it, like how objects are structured or how they interact with each other. CNNs are good at understanding an image from various three-dimensional views, but they aren't so great at understanding how it relates to other images in the same way humans do. To understand capsule networks or what Hinton has called "dynamic routing between capsules," it is essential to comprehend CNNs. Let's use some visual examples to get a sense of what this means. CNNs are good at understanding an image from various three-dimensional views. Still, they could better understand how it relates to other images in the same way humans do. For example, if you point your camera at a table lamp and take a photo, a CNN will give you pretty accurate results for the lamp itself. It can identify its color and shape. But when it comes to an understanding what kinds of things belong on a table (like books) or what types of things belong on a nightstand (like alarm clocks), CNNs still need to be more effective at making those connections. The idea that a computer can understand objects and relationships by breaking them down into three-dimensional pieces is familiar. It's a concept that's been around for decades. The idea that you could do this on a massive scale (up to billions of objects) and do it quickly enough to make it worthwhile, well… that's just crazy talk. But as we all know, crazy ideas are often the ones that change the world. Hinton's approach to AI has led to groundbreaking research into more powerful AI systems.