- ホーム
- > 洋書
- > 英文書
- > Computer / General
Full Description
Learn the mathematics behind large language models and deep learning, from optimization and attention to diffusion, uncertainty, scaling laws, and hallucination, with equations explained symbol by symbol from first principles.
Key Features
Connect core mathematics to transformers, diffusion, and modern generative models
Understand optimization, generalization, uncertainty, robustness, and causality
Trace AI theory from foundational notation to scaling laws, hallucination, and abstention
Book DescriptionModern AI spans more than neural-network architectures: understanding why models learn, generalize, generate, scale, and fail requires the mathematics connecting them. This book builds that foundation systematically, starting with notation, approximation and expressivity, optimization and training dynamics, generalization, symmetry, and geometric inductive bias.
Written by Jason Karpeles, a former world No. 5-ranked Kaggle competitor with 30 years of predictive-modeling experience, the book keeps important equations and derivations intact. For each key formula, it explains both what the equation does and how to read its notation symbol by symbol. You'll then examine convolutional and multiscale models, recurrent and state-space models, attention and transformers, graph and geometric deep learning, and generative approaches spanning latent-variable, flow, diffusion, and score-based models.
Later chapters cover neural ODEs, scientific machine learning, Bayesian deep learning, uncertainty, robustness, out-of-distribution behavior, causality, scaling laws, and in-context learning. The final chapter examines truth, error, abstention, and hallucination as mathematical problems. By connecting theory with clear interpretation, the book helps you follow AI research, compare model families, and reason precisely about model behavior.What you will learn
Read linear algebra, calculus, and probability notation used in AI
Trace gradient descent and backpropagation through model training
Explain approximation, generalization, and implicit bias in networks
Derive attention and transformer equations from first principles
Compare CNNs, RNNs, state-space, graph, and geometric models
Follow the math of VAEs, flows, diffusion models, and neural ODEs
Quantify uncertainty, robustness, OOD behavior, and causality
Interpret scaling laws, in-context learning, and LLM hallucination
Who this book is forThis book is for machine learning engineers, data scientists, AI researchers, quantitative practitioners, technical students, and mathematically curious readers who want to understand the theory behind modern deep learning and large language models. It is especially relevant to readers who need to follow derivations, interpret research papers, or reason about model behavior beyond high-level intuition. Familiarity with basic linear algebra, calculus, and probability is helpful, although the opening chapters review core notation and concepts. No programming is required.
Contents
Table of Contents
Foundations and Notation
Approximation and Expressivity
Optimization and Training Dynamics
Generalization and Implicit Bias
Symmetry and Geometric Inductive Bias
Convolutional and Multiscale Models
Recurrent and State-Space Model
Attention and Transformer
Graph and Geometric Deep Learning
Latent, Adversarial, and Flow Models
Diffusion, Score, and Transport-Based Generation
Continuous-Depth and Implicit Models
Scientific ML and Operator Learning
Bayesian Deep Learning and Uncertainty
Robustness, OOD, and Causality
Scaling Laws, In-Context Learning, and Abstract Foundations
The Mathematics of Truth, Error, and Abstention



