Convolutional Neural Networks (CNN)

Coming soon — this chapter will cover the core ideas behind convolutional neural networks, the architectures that drove the deep learning revolution in computer vision.

Overview

Convolutional Neural Networks (CNNs) are a family of deep neural networks designed to process data with a grid-like topology — most notably images. By exploiting local connectivity, weight sharing, and translation equivariance, CNNs learn hierarchies of features that progress from edges and textures to objects and scenes.

What you’ll learn

• The convolution operation, kernels, stride, padding, and receptive fields
• Pooling, normalization, and activation functions in vision networks
• Classic architectures: LeNet, AlexNet, VGG, GoogLeNet/Inception, ResNet
• Modern variants: DenseNet, MobileNet, EfficientNet, ConvNeXt
• Beyond classification: detection, segmentation, and dense prediction
• Practical training: data augmentation, transfer learning, regularization

Why it matters

CNNs are the workhorse of computer vision and remain a strong baseline for many perception tasks even in the era of Transformers. Understanding their inductive biases — locality and translation equivariance — is essential to choosing the right model and reasoning about what a network can and cannot learn from limited data.