Librería Samer Atenea
Kálamo Books
Librería Elías (Asturias)
Librería Kolima (Madrid)
Librería Proteo (Málaga)
Gradient of Forgetting explores one of modern artificial intelligence’s most consequential challenges: catastrophic forgetting. As neural networks are repeatedly fine-tuned, updated, and deployed, valuable capabilities can quietly disappear. Yet sometimes forgetting is exactly what we need, especially when sensitive information must be removed.Written for researchers, students, and AI engineers, this practical guide explains why forgetting occurs, how to measure it accurately, and how to manage the trade off between preserving knowledge and learning new tasks. It compares four major approaches regularisation, replay, parameter isolation, and optimisation without overlooking their memory, computational, and privacy costs.Across 21 focused chapters, the book connects foundational theory to contemporary challenges in pretrained models, language models, streaming systems, federated learning, and machine unlearning. Realistic failure scenarios, reproducible evaluation methods, deployment safeguards, and governance checklists show how to build systems that remain dependable as they evolve.With clear explanations, hands-on Python and PyTorch examples, practical exercises, and a companion code repository, Gradient of Forgetting offers the tools to evaluate continual-learning methods critically and create adaptive AI systems that know what to retain, what to update, and what to erase.