Volusia discussed on Software Engineering Daily

Automatic TRANSCRIPT

What you would also want to have something that might fire on the ears of the cat or maybe on the body or on the tail of the cat SE. This is this is basically gradually. It's a gradual process would start looking at very very narrow and and local features as we called them. And as you basically at more and more of these layers, you will try to explain the whole cat by means of its individual pieces. If you want by assembling, basically progress to more and more people layers in over network. So if we were to take a matrix that was a ten by ten matrix. And it just had ones and zeroes in it you, and I would be able to identify. Hey, look in the upper right hand corner of that ten by ten matrix. There's a bunch of ones up there. And and we could agree on that. We could say look there happens to be a lot of ones up in that right hand corner in what this conflusion system of finding. Edges or finding circles is doing is very similar. It's just doing that at a deeper more complex way where computer can say, you know, much like you, and I could could with our basic analysis could say there's a bunch of ones in the upper right hand corner. A computer could say this collection of pixels in this space over here. This looks like a circle or this looks like an eye, and then you can start to build up these higher levels of abstraction that are associated with these matrices of pixels. Guess so we'll do we'll be able to identify. It's basically those come Volusia kernel parameters there eventually being learned through this process of training, a deep neural network or or or not working in that simply sample here. So you would actually try to write down your task as a kind of up to my station problem. And the optimization problem would be solved in a way that you basically try to find what would an individual convoluted operator have to look like in order to best possibly explain or self optimization problem. And if it's something that would have to sort of respond to the eye of a cat in the best possible way to solve your problem. Because all you want to do is finding is of cats than that, Volusia Colonel would look probably very much like an eye of a cat. If will be possible. So what would really try to do is find exactly the kind of pattern? The best describes your data that you use for learning. This basically your training data that you have in your your loaning problem. Right. Yeah. You might have a collection of images of cats and somebody has gone through those images and labeled the eyes of the cats. So that the computer can idea. Defy that this section of the picture is a cat. I there's a number of other terms, we could we could go through..

Coming up next