Learning decision sequences for repetitive processes--selected algorithms

Contents
List of Figures
List of Tables
1 Introduction
2 Basic Notions and Notations
	2.1 Repetitive Processes
		2.1.1 Process States
		2.1.2 States of Repetitive Processes
	2.2 Decision Sequences and Disturbances
		2.2.1 Univariate Decision Sequences
		2.2.2 Multivariable Decision Sequences
		2.2.3 Random Disturbances
		2.2.4 Decision Making and Implementation
	2.3 Static Models of Processes and Loss Functions
		2.3.1 Model-Based Versus Model-Inspired Approaches
		2.3.2 Deterministic Static Models
		2.3.3 Probabilistic Static Models
		2.3.4 Assessing the Quality of Decision Sequences
		2.3.5 Illustrative Example
	2.4 Models of Dynamic Processes
		2.4.1 Markov Chains Models
		2.4.2 Deterministic Models and Miscellaneous Remarks
		2.4.3 Quality Criteria for Dynamic Processes
3 Learning Decision Sequences and Policies for Repetitive Processes
	3.1 Learning Decisions and Decision Sequences
		3.1.1 Remarks on Learning in Control Systems
		3.1.2 Selected Learning Tasks in Operations Research
	3.2 How Algorithms Can Learn
		3.2.1 Learning directly from a static process I
		3.2.2 Learning by stochastic approximation
		3.2.3 Remarks on model free learning
		3.2.4 Possible Roles of Models in Learning
	3.3 Plug-In Versus the Nonparametric Approach to Learning
4 Differential Evolution with a Population Filter
	4.1 Filter
	4.2 Filter in Global Optimization Problems
		4.2.1 Evolutionary Computations
		4.2.2 Differential Evolution
5 Decision Making for COVID-19 Suppression
	5.1 Modified Logistic Growth Model and Its Validation for Poland
		5.1.1 Classic Logistic Growth Models
		5.1.2 Modified Logistic Growth Model
		5.1.3 Bernstein Polynomials as Possible Models of the Epidemic Growth Rate
		5.1.4 Model Discretization and Validation
	5.2 Optimization of Decision Sequence—Problem Statement
		5.2.1 Actions Reducing the Spread Of COVID-19
		5.2.2 Constraints
		5.2.3 Interpreting the Goal Function
	5.3 Searching for Decisions Using Symbolic Calculations
		5.3.1 Model-Based Prediction by Symbolic Calculations
		5.3.2 The Newton Method Using Hybrid Computations
		5.3.3 Testing Example—COVID-19 Mitigation
	5.4 Learning Decisions by Differential Evolution with Population filter
		5.4.1 Solving the Optimization Problem
		5.4.2 Robustness of Differential Evolution
6 Stochastic Gradient in Learning Decision Sequences
	6.1 Model-Free Classic Approach—Stochastic Approximation Revisited
		6.1.1 Stochastic Approximation—Problem Statement
		6.1.2 The Kiefer–Wolfowitz Algorithm
		6.1.3 Modifications of the Kiefer–Wolfowitz Algorithm
		6.1.4 Generalizations of the K-WSAA for Handling Constraints
	6.2 Random, Simultaneous Perturbations for Estimation Gradient at Low Cost
		6.2.1 The Idea of Simultaneous Random Perturbations
		6.2.2 Simultaneous Perturbation Algorithm for Decision Learning (SPADL)
	6.3 Response Surface Methodology for Searching for the Optimum
		6.3.1 Gradient Estimation According to Response Surface Methodology
		6.3.2 RS Methodology—Learning Algorithm
		6.3.3 Selecting Experiment Designs
	6.4 Discussion on Stochastic Gradient Descent Approaches
		6.4.1 Selecting the Step Length and Scaling
7 Iterative Learning of Optimal Decision Sequences
	7.1 Run-to-run Control as an Inspiration
		7.1.1 Learning in Run-to-run Decision Systems
		7.1.2 Outline of the Run-to-run Optimization Algorithm
	7.2 Iterative Learning Control—In Brief
		7.2.1 Basic Formulation of the ILC
		7.2.2 An Optimization Paradigm in ILC Theory
	7.3 Iterative Learning of Optimal Decision Sequences
	7.4 Derivation of the Learning Algorithm
	7.5 Pass to Pass Learning
8 Learning from Image Sequences
	8.1 Motivation and Aims
	8.2 Proposed Classifier
	8.3 Example
Appendix  Bibliography
Index