Mathematics in Deep Learning. CNNs, Transformers, Diffusion Models, and LLMs (en Inglés) - Yin Yang

Mathematics in Deep Learning is a practical textbook for readers who have seen neural networks in code and want the mathematics behind them to feel usable, explanatory, and connected to practice.

The book builds the habits that make deep learning easier to reason about: tracking tensor shapes, reading models as parameterized functions, connecting losses to data, understanding optimization, and checking whether learned rules will behave well away from the training examples.

Topics move from foundations to modern systems, including tensors, probability, empirical risk, convolution, backpropagation, CNN architecture, transfer learning, embeddings, sequence models, attention, transformers, autoregressive modeling, reinforcement learning, preference optimization, training theory, vision-language models, object detection, segmentation, generative modeling, speech recognition, and speech generation.

Each chapter uses running examples, compact derivations, figures, exercises, and companion SymPy or numerical code to keep the mathematics tied to inspectable computations.

For students, engineers, and self-study readers who want to understand deep learning models more clearly, not just run them.