Convolutional Networks

• Convolution operation: filters, stride, padding, receptive field
• Pooling: max pooling, average pooling, global average pooling
• Batch normalisation, dropout, data augmentation
• Landmark architectures: LeNet, AlexNet, VGG, GoogLeNet/Inception, ResNet (skip connections), DenseNet
• Efficient architectures: MobileNet (depthwise separable convolutions), EfficientNet (compound scaling), ShuffleNet
• Transfer learning: feature extraction, fine-tuning pretrained backbones
• Visualising CNNs: activation maps, Grad-CAM, feature inversion