Convolutional Neural Networks
As models grow larger and more complex, the number of parameters explodes, making model optimization slow and computationally expensive. However, the number of parameters can sometimes be reduced because neighboring parts of grid-structured data, for example neighboring pixels in an image, often contain similar information. Convolutional neural networks (CNNs) exploit this property by convolving the data with different kernels, also called filters, to aggregate the related information before being processed by the network’s dense layers. The different kernels can also serve as detectors that identify salient features in the data and pass those on to the next layer in the model.
Through interactive demos and hands-on coding exercises, you will experiment with kernel size, stride, and more to get an intuition for how convolution works, learn how to create a convolutional neural network, and apply CNNs to classify images.