8 Using convolutions to generalize
This chapter covers
- Understanding convolution
- Building a convolutional neural network
- Creating custom
nn.Modulesubclasses - The difference between the module and functional APIs
- Design choices for neural networks
In the previous chapter, we built a simple neural network that could fit (or overfit) the data, thanks to the many parameters available for optimization in the linear layers. Our model used fully connected layers, where each neuron connects to every neuron in the previous layer, treating the image as a flattened vector with no spatial structure preserved. However, we had issues with our model in that it was better at memorizing the training set than it was at generalizing the properties of birds and airplanes. Based on our model architecture, we’ve got a guess as to why that’s the case. Due to the fully connected setup needed to detect the various possible translations of the bird or airplane in the image, we have both too many parameters (making it easier for the model to memorize the training set) and no position independence (making it harder to generalize). As we discussed in the last chapter, we could augment our training data by using a wide variety of re-cropped images to try to force generalization, but that won’t address the issue of having too many parameters.