Convolutional Neural Networks

Image recognition, feature hierarchies, and the architectures that transformed computer vision. This lecture covers the mathematical foundations of convolution operations, pooling strategies, and how deep convolutional networks learn hierarchical representations from raw pixel data.

Topics include LeNet, AlexNet, VGGNet, ResNet, and modern efficient architectures. We examine how skip connections solved the degradation problem, enabling networks with hundreds of layers, and discuss the transition from hand-crafted features to learned representations.