This chapter offers a comprehensive exposition of the fundamental theories and key techniques of deep learning, encompassing its core concepts, foundational models, learning objectives, optimization strategies, and representative architectures. It begins by introducing the historical context and the central issue of contribution allocation, using artificial neural networks to illustrate the information flow and their advantages in representation learning over traditional methods. The training pipeline is then examined in detail, including model formulation, loss function design, gradient-based optimization, regularization, and early stopping, providing a unified view of deep learning from both theoretical and practical perspectives. Structurally, the chapter highlights the principles of local connectivity, weight sharing, and pooling in convolutional neural networks, with classic architectures used to illustrate their evolution. Finally, this chapter introduces variational autoencoders and generative adversarial networks, discussing their modeling frameworks and application scenarios.