現代優化方法 (英文) 李慶娜 9787030747853 【台灣高等教育出版社】

精彩書評
本書是作者多年研究工作的總結

目錄

Contents
Preface III
CHAPTER 1
Introduction 1 1
1 1 About Optimization1
1 2 Classification of Optimizatiou4
1 3 Preliminaries in Convex Analysis10
1 4 Exercises15
CHAPTER 2
Fundamentals of Optimization 17
2 1 Unconstrained Optimization Problem17
2 2 What is a Solution 18
2 2 1 Definitions of Different Solutions18
2 2 2 Recognizing a Local Minimun20
2 2 3 Nonsmooth Problems23
2 3 Overview of Algorithms25
23 1 Line Search Strategy26
2 3 2 Trust Region Strategy30
2 4Couvergence31
2 5Scaling32
2 6Exercises 33
CHAPTER 3
Line Search Methods35
3 1Step Length35
3 1 1 The Wolfe Conditions37
3 1 2 The Goldstein Conditions40
3 1 3 Suficient Decrease and Backtracking41
3 2 Convergence of Line Search Methods42
3 3 Rate of Convergence44
3 3 1 Steepest Descent Method44
3 3 2 Newton's Method16
3 3 3 Quasi-Newton Methods48
3 4 Exercises50
CHAPTER 4
Trust Regiou Methods51
4 1Outline of the Trust Region Approach52
4 2Algorithms Based on the Cauchy Point 54
4 2 1 The Caucly Poiut54
4 2 2 The Dogleg Method56
4 2 3 Two-Dimensioual Subepace Minimization 58
4 3Global Convergence59
4 3 1 Reduction Obtained by the Caucly Point59
4 3 2 Couvergence to Stationary Points 61
4 4 Local Convergence65
4 5 Other Enhancements 46
4 6 Exercises68
CHAPTER 5 Comjugate Gradient Methods 69
5 1 Linear Conjugate Gradient Method 69
5 1 1 Coujugate Direction Method 69
5 1 2 Conjugate Gradient Method 72
5 1 3 A Practical Form of the Conjugate Gradient Method 75
5 1 4 Rate of Couvergence 76
5 1 5 Precouditioning 77
5 2 Noulinear Conjugate Gradient Methods78
5 2 1 The Polak- Ribiere Method and Variants30
5 2 2 Global Convergence81
5 3 Exercises 83
CHAPTER 6
Seamismooth Newton's Method 85
6 1 Semismoothuness85
6 2 Nousmooth Versiou of Newton's Method 87
6 3 Support Vector Machine 89
6 4 Semismooth Newton's Method for SVM 91
6 5 Exercises 96
CHAPTER 7
Theory of Constrained Optimization 97
7 1 Local and Global Solutions 97
7 1 1 Smoothmess 98
7 2 Examples99
7 3 Tangent Cone and Constraint Qualifications 103
7 4 First- Order Optimality Conditious 105
7 5 Second- Order Conditions106
7 6 Duality109
7 7 KKT Condition 112
7 8 Dual Problem 114
7 9 Exercises 118
CHAPTER 8
Penalty and Augmented Lagrangian Methods 119
8 1 The Quadratic Penalty Method 119
8 2 Exact Penalty Method 122
8 3 Augmented Lagrangian Method 123
8 4 Quadratic Penalty Method for Hypergaph Matching 125
8 4 1 Hypergraph Matching 126
8 4 2 Mathematical Formulation 126
8 4 3 Relaxation Problem 128
8 4 4 Quadratic Penalty Method for (8 21)129
8 4 5 Numerical Results 130
8 5 Augmented Lagrangian Method for SVM 132
8 5 1 Support Vecotr Machine 132
8 5 2 Mathermatical Formulation 133
8 5 3 Augmented Lagrangian Method (ALM) 133
8 5 4 Sermismooth Newton's Method for the Subproblem 136
8 5 5 Reducing the Computational Cost 137
8 5 6 Convergence Result of ALM 138
8 5 7 Numerical Results on LIBLINEAR 139
8 6 Exercises 141
CHAPTER 9
Bilevel Optimization and Its Applicatious143
9 1 Introduction 143
9 2 Bilevel Model for a Case of Hyperparameter Selection in SVC 145
9 2 1 An MPEC Formulation 147
9 3 The Global Relaxation Method (GRM)148
9 4 MPEC-MFCQ: A Hidden Property 149
9 5 Numerical Resutlts 150
Bibliography 153
↑

精彩書摘
Chapter 1 Introduction
In the first chapter of this book， we will introduce what is optimization， the classification of optimization， 8a8 well a8 some preliminaries in conuvex analysis
1 1 About Optimization
Optimization exists everywhere
People optinize As long a pople have choces， they do optimization In finance， people do portfolio eelectiong to maximize the rate of return and avoid risk
In engineering， people try to maximize the efficiency of the system and try to optimal control of the system
Nature optimizes as well Physical systems always reach the state of total minimum energy For isolated chemical systems reactions will not stop until reaching the minimum total potential energy Light travels following the path of minimizing travel time See figure 1 1 Figure 1 1， S represents the incident ray， and S' represents the mirror ray of 8 with respect to the reflective surface
Fia 1 1 - Light travels fllowins the shortest path
One can conclude that 88 long as there are choices， optimization happens
There are three key facts in optimization () Objective It is a quantitative measure of the performnance under study For example， profit， time， and potential energy (l) Variables They are the unknowns in the problem， which need to be determined (ii) Coustraints They are the restrictions tbat variables follow， such as nonnegativity， and 50 ou
The optimization process can be divided into the fllowring three steps
[i) Modelling It is to identify the key facts in optimization Ou one hand， the mode! can not be two simple If the rmodel is t0o simple， it will not represent the real application problems On the other hand， it can not be two coupli-cated， because it will bring challenges in 8olving the problem
(ii) Apply optimization algorithm to solve the model This