Foundations and Engineering of Artificial intelligence: Mathematical Foundations, Deep and Reinforcement Learning, and Large Language Models
English | October 7, 2026 | ASIN: B0HM66PXX1 | 20 Pages | Epub | 35.07 MB
Artificial intelligence is transforming science, engineering, communication, healthcare, robotics, education, industry, and everyday life. Yet understanding modern AI requires more than learning how to call an existing model through an API. It requires understanding the mathematical principles, learning algorithms, neural architectures, optimization methods, and computational processes that make intelligent systems possible.This book provides a progressive introduction to the foundations of contemporary artificial intelligence, connecting mathematics, machine learning, deep learning, reinforcement learning, large language models, and interactive AI systems within a unified framework.The book begins with the mathematical foundations required to understand modern AI, including linear algebra, probability, statistics, information theory, calculus, gradients, Jacobians, Hessians, and optimization. These concepts are then used to develop neural networks from the artificial neuron upward, covering activation functions, multilayer perceptrons, forward propagation, backpropagation, normalization, regularization, residual connections, convolutional neural networks, and recurrent architectures.Reinforcement learning is developed from first principles through agents, environments, states, actions, rewards, policies, value functions, Markov decision processes, Bellman equations, exploration, temporal-difference learning, SARSA, and Q-learning. The discussion then advances to deep reinforcement learning, including Deep Q-Networks, policy-gradient methods, actor-critic algorithms, and Proximal Policy Optimization.A major part of the book is devoted to the foundations and engineering of large language models. Readers are guided through tokenization, embeddings, positional representations, query-key-value projections, causal multi-head self-attention, masking, feed-forward networks, normalization, residual connections, Transformer architectures, next-token prediction, and autoregressive generation.The book also examines modern techniques for improving model capability and efficiency, including rotary position embeddings, RMSNorm, SwiGLU, grouped-query attention, multi-query attention, latent attention mechanisms, and mixture-of-experts architectures.The final chapters explain how pretrained language models are adapted into more capable AI systems through prompting, zero-shot and few-shot learning, supervised fine-tuning, LoRA, QLoRA, preference learning, reward modeling, reinforcement learning from human feedback, Direct Preference Optimization, inference-time reasoning, retrieval, external tool use, safety, behavioral alignment, and multi-step interaction.Throughout the book, mathematical expressions are connected directly to algorithms and computational procedures. The emphasis is not only on what AI systems can do, but on how information is represented, how model parameters are organized, how learning objectives are constructed, how gradients update those parameters, how attention operates, how Transformers generate tokens, and how trained models become components of larger intelligent systems.Designed for upper-level undergraduate and graduate students, engineers, researchers, and technical professionals, this book is suitable for readers in computer science, electrical and computer engineering, data science, robotics, communications, and related disciplines.Its central goal is to provide the mathematical, algorithmic, and engineering foundations needed to understand how modern AI systems are built, trained, adapted, evaluated, and deployed, while preparing readers to contribute to the development of the next generation of artificial intelligence.
Links are Interchangeable - Single Extraction
