The Evolution of Neural Networks

The algorithm is actually represented by the structure of the machine like what what nodes are connected to what. And you start out with a machine that's just basically randomly built. It's as like, you know, but you show it a cat, it has a one in three chance of saying cat. But then you fiddle with one note we change one connection and see if the performance gets better or worse. If it improves the performance you leave it the way you fiddled with it if it makes it worse you blow up. You just do that forever until the machine is getting things almost 100% right. That's why they call it stochastic gradient descent because you're feeding it a