Google Inceptions, Neural Networks. AKA The Dog-Fish

This Google Research blog post is a remarkable look at the work they been doing in training Artificial Neural Networks to recognize images (aka Google Images). That feat in and of itself is rather stunning but this article goes on to explain some of the mishaps and complexities of how a network can see things that aren't there. It's not unlike how we mistake clouds for particular shapes, except with a computer envisioning those things you actually get to see inside what it's seeing. It's basically the equivalent of visualized imagination.

If you're not into reading, just check out this gallery and marvel at the Acid Trip style creations the networks make when asked to describe an image multiple times.

via Chad