 So what do we have so far? It's not much of a hierarchy. We have a convolution, a reload, a pooling, giving us the results that you can see on the right-hand side. But now, how do we build actual networks out of this? Well, we can chain that, of course. We take a convolution, a reload, another convolution, another reload, a pooling, a convolution, a reload, a pooling. Here you see the strategy. Here, we build convolution, reload, convolution, reload. Instead of convolution, reload, pooling, convolution, reload, pooling. Why? Because this leads to fewer parameters, which is in a lot of ways desirable. So what would the output of that be? Well, in this case, it might be that for one feature, we would get this, for another, we'd get this, for the third one, we'd get this. And now, ultimately, we can have a fully connected part that converts these outputs of the conf layers into, that flattens these, and that ultimately then does the readout and tells us which class that is. Now, we want to predict both, of course, how likely the X is and how likely the O is. And we'd be getting activations for that. Now, we might want that the probabilities for the two of them add up to one, and we are already hot about cross entropy. And now, of course, it usually helps to have multiple layers of fully connected. You can say, I want to take the output of the conf layers, I want to flatten it, and then I want to apply multiple fully connected layers to that. And now, we can stake all of that, convolution, reload, convolution, reload, pooling, convolution, reload, pooling, fully connected, fully connected, and then we have the output. Now, why are we doing that? Well, because it's a good way of building that. So what I want you to do is understand the full X versus O network that we built and construct it, and really grok how we construct such networks. Play with various ways how it can run, and we will train it. But ignore the training part of that. Not like we will, in the next tutorial, we will hear about how we actually train such networks. So while this is the network that's being trained, I just want you to be able to construct the conf net, because next week we will be talking a lot about how we can train conf nets. So now, go and understand the typical conf net and construct it.