Brain and Nature-Inspired Learning, Computation and Recognition

Licheng Jiao, Ronghua Shang, Fang Liu, Weitong Zhang

• Publisher: Elsevier
• Year: 2020
• Print ISBN: 9780128197950
• eBook ISBN: 9780128204047
• Topic: Technology & Engineering
• Subtopic: Engineering General
• Index: Technology & Engineering

Brain and Nature-Inspired Learning, Computation and Recognition presents a systematic analysis of neural networks, natural computing, machine learning and compression, algorithms and applications inspired by the brain and biological mechanisms found in nature. Sections cover new developments and main applications, algorithms and simulations. Developments in brain and nature-inspired learning have promoted interest in image processing, clustering problems, change detection, control theory and other disciplines. The book discusses the main problems and applications pertaining to bio-inspired computation and recognition, introducing algorithm implementation, model simulation, and practical application of parameter setting. Readers will find solutions to problems in computation and recognition, particularly neural networks, natural computing, machine learning and compressed sensing. This volume offers a comprehensive and well-structured introduction to brain and nature-inspired learning, computation, and recognition. - Presents an invaluable systematic introduction to brain and nature-inspired learning, computation and recognition - Describes the biological mechanisms, mathematical analyses and scientific principles behind brain and nature-inspired learning, calculation and recognition - Systematically analyzes neural networks, natural computing, machine learning and compression, algorithms and applications inspired by the brain and biological mechanisms found in nature - Discusses the theory and application of algorithms and neural networks, natural computing, machine learning and compression perception

Chapter 1

Introduction

Abstract

In recent years, brain- and nature-inspired algorithms have emerged endlessly. Neural network, natural computing, machine learning, and compressed sensing have achieved great success in image processing, clustering, and change detection, especially in SAR image segmentation, polarization synthetic aperture radar (SAR) segmentation, and community detection. Brain- and nature-inspired learning computation and recognition are research directions that have attracted wide attention in recent years. They have become the main component of AI 2.0 in China. In this book, theoretical knowledge and specific applications are discussed and analyzed in detail, to meet the dual needs of academic research and engineering innovation. This book systematically discusses the basic theories, algorithms, and applications of neural networks, natural computing, machine learning, and compressed sensing, so as to facilitate further research and exploration by interested readers.

Keywords

Brief introduction; Compressive sensing; Machine learning; Neural network

1. 1.1 A brief introduction to the neural network
   1. 1.1.1 The development of neural networks
   2. 1.1.2 Neuron and feedforward neural network
   3. 1.1.3 Backpropagation algorithm
   4. 1.1.4 The learning paradigm of neural networks
2. 1.2 Natural inspired computation
   1. 1.2.1 Fundamentals of nature-inspired computation
   2. 1.2.2 Evolutionary algorithm
   3. 1.2.3 Artificial immune system (AIS)
   4. 1.2.4 Other methods
3. 1.3 Machine learning
   1. 1.3.1 Development of machine learning
   2. 1.3.2 Dimensionality reduction
   3. 1.3.3 Sparseness and low-rank
   4. 1.3.4 Semisupervised learning
4. 1.4 Compressive sensing learning
   1. 1.4.1 The development of compressive sensing
   2. 1.4.2 Sparse representation
   3. 1.4.3 Compressive observation
   4. 1.4.4 Sparse reconstruction
5. 1.5 Applications
   1. 1.5.1 Community detection
   2. 1.5.2 Capacitated arc routing optimization
   3. 1.5.3 Synthetic aperture radar image processing
   4. 1.5.4 Hyperspectral image processing
6. References

1.1. A brief introduction to the neural network

Over the years, scientists have been exploring the secrets of the human brain from various perspectives, such as medicine, biology, physiology, philosophy, computer science, cognition, and organization synergetics, hoping to make artificial neurons that simulate the human brain. In the process of research, in recent years, a new multidisciplinary cross-technology field has been formed, called “artificial neural network” The research into neural networks involves a wide range of disciplines, which combine, infiltrate, and promote each other.

Artificial neural network (ANN) is an adaptive nonlinear dynamic system composed of a large number of simple basic elements—neurons. The structure and function of each neuron are relatively simple, but the system behavior produced by a large number of neuron combinations is very complex. The basic structure of an artificial neural network mimics the human brain, and reflects some basic characteristics of human brain function. It can adapt itself to the environment, summarize rules, and complete some operations, recognition, or process control. Artificial neural networks have the characteristics of parallel processing, which can greatly improve work speed.

1.1.1. The development of neural networks

The development of artificial neural networks has gone through three climaxes: control theory from the 1940 to 1960s [1-3], connectionism from the 1980s to the mid-1990s [4, 5], and deep learning since 2006 [6, 7].

In 1943, Warren McCulloch and Walter Pitts based on a mathematical algorithm threshold logic algorithm created a neural network model [8]. This linear model identifies two different types of inputs by testing whether the response output is positive or negative. The study of neural networks is divided into the study of biological processes in the brain and the study of artificial intelligence (artificial neural networks). In 1949, Hebb published Organization of Behavior, and put forward the famous “Hebb theory” [2] Hebb theory mainly argues that when the axons of neuron A are close to neuron B and neuron A participates in the repeated and sustained excitement of neuron B, both the neurons or one of them will change the process of growth or metabolism, which can enhance the effectiveness of neuron A stimulating neuron B [9]. Hebb theory was confirmed by Nobel Prize winner Kendall and his animal experiments in 2000 [10]. The later unsupervised machine learning algorithms are the variants of Hebb theory more or less. In 1958, Frank Rosenblatt simulated a neural network model called the “perceptron” which was invented on an IBM-704 computer [11]. This model can perform some simple visual processing tasks. Rosenblatt believed that the perceptron would eventually be able to learn, make decisions, and translate languages. In 1959, another two American engineers, Widrow and Hoff [12], put forward the adaptive linear element (Adaline). This was a change from the perceptron and one of the progenitor models of machine learning. The main difference between it and perceptron was that the Adaline neuron has a linear activation function, which allows the output t...

Table of contents

• Cover image
• Title page
• Table of Contents
• Copyright
• Chapter 1. Introduction
• Chapter 2. The models and structure of neural networks
• Chapter 3. Theoretical basis of natural computation
• Chapter 4. Theoretical basis of machine learning
• Chapter 5. Theoretical basis of compressive sensing
• Chapter 6. Multiobjective evolutionary algorithm (MOEA)-based sparse clustering
• Chapter 7. MOEA-based community detection
• Chapter 8. Evolutionary computation-based multiobjective capacitated arc routing optimizations
• Chapter 9. Multiobjective optimization algorithm-based image segmentation
• Chapter 10. Graph-regularized feature selection based on spectral learning and subspace learning
• Chapter 11. Semisupervised learning based on nuclear norm regularization
• Chapter 12. Fast clustering methods based on learning spectral embedding
• Chapter 13. Fast clustering methods based on affinity propagation and density weighting
• Chapter 14. SAR image processing based on similarity measures and discriminant feature learning
• Chapter 15. Hyperspectral image processing based on sparse learning and sparse graph
• Chapter 16. Nonconvex compressed sensing framework based on block strategy and overcomplete dictionary
• Chapter 17. Sparse representation combined with fuzzy C-means (FCM) in compressed sensing
• Chapter 18. Compressed sensing by collaborative reconstruction
• Chapter 19. Hyperspectral image classification based on spectral information divergence and sparse representation
• Chapter 20. Neural network-based synthetic aperture radar image processing
• Chapter 21. Neural networks-based polarimetric SAR image classification
• Chapter 22. Deep neural network models for hyperspectral images
• Index