Deep learning models are powerful, but often large, slow, and expensive to run. This book is a practical guide to accelerating and compressing neural networks using proven techniques such as quantization, pruning, distillation, and fast architectures. It explains how and why these methods work, fostering a comprehensive understanding. Written for engineers, researchers, and advanced students, the book combines clear theoretical insights with hands-on PyTorch implementations and numerical results. Readers will learn how to reduce inference time and memory usage, lower deployment costs, and select the right acceleration strategy for their task. Whether you're working with large language models, vision systems, or edge devices, this book gives you the tools and intuition needed to build faster, leaner AI systems, without sacrificing performance. It is perfect for anyone who wants to go beyond intuition and take a principled approach to optimizing AI systems

‘This book is a practical guide to DNN and LLM acceleration, bridging the gap between theory and practice. Moving beyond ‘black-box’ tricks, it pairs the latest techniques-like FlashAttention-with runnable code and empirical data. Readers will gain both the technical tools and the fundamental understanding to optimize models effectively.’

Masashi Sugiyama - RIKEN and University of Tokyo

‘This book effectively bridges theory and practice in accelerating deep learning. It offers clear insights into modern architectures such as Mamba, while also elucidating fundamental concepts and practical techniques for efficient deep learning. It will be a valuable resource for researchers and graduate students seeking a deep understanding of modern deep learning.’

Makoto Yamada - Okinawa Institute of Science and Technology