From theory to application, this book presents research on biologically and brain-inspired networking and machine learning based on Yuragi, which is the Japanese term describing the noise or fluctuations that are inherently used to control the dynamics of a system. The Yuragi mechanism can be found in various biological contexts, such as in gene expression dynamics, molecular motors in muscles, or the visual recognition process in the brain. Unlike conventional network protocols that are usually designed to operate under controlled conditions with a predefined set of rules, the probabilistic behavior of Yuragi-based control permits the system to adapt to unknown situations in a distributed and self-organized manner leading to a higher scalability and robustness.
The book consists of two parts. Part 1 provides in four chapters an introduction to the biological background of the Yuragi concept as well as how these are applied to networking problems. Part 2 provides additional contributions that extend the original Yuragi concept to a Bayesian attractor model from human perceptual decision making. In the six chapters of the second part, applications to various fields in information network control and artificial intelligence are presented, ranging from virtual network reconfigurations, a software-defined Internet of Things, and low-power wide-area networks.
This book will benefit those working in the fields of information networks, distributed systems, and machine learning who seek new design mechanisms for controlling large-scale dynamically changing systems.
Author(s): Masayuki Murata, Kenji Leibnitz
Publisher: Springer
Year: 2021
Language: English
Pages: 247
City: Singapore
Preface
Contents
Contributors
Part I Fluctuation-Based Control Systems: Yuragi Concept
1 Introduction to Yuragi Theory and Yuragi Control
1.1 Introduction
1.2 Principles of Self-organization
1.2.1 Centralized and Distributed Control
1.2.2 Characteristics of Self-organized Systems
1.2.3 Role of Noise in Self-organized Systems
1.3 Examples of Nature-Inspired Models Utilizing Noise
1.3.1 Random Walks and Brownian Motion
1.3.2 Impact of Noise on Visual Perception and Decision-Making in the Brain
1.3.3 Signal Enhancement Through Stochastic Resonance
1.3.4 Evolutionary and Genetic Algorithms
1.3.5 Routing Methods Inspired by Social Insects
1.3.5.1 Ant Colony Optimization (ACO)
1.3.5.2 AntNet in Packet-Switched Networks
1.3.5.3 AntHocNet in Mobile Ad-Hoc Networks
1.3.5.4 BeeHive for Wired Connectionless Networks
1.3.6 Tug-of-War Model for Solving the Multi-Armed Bandit Problem
1.4 Mathematical Formulation of Noise-Driven Systems
1.4.1 Stability and Attractors
1.4.2 Dynamic Systems Under the Influence of Noise
1.4.3 Relationship Between Fluctuation and Its Response
1.5 Yuragi Model for Attractor Selection
1.5.1 Adaptive Response of Gene Network to Nutrient Availability
1.5.2 Modeling the Interactions of Gene Expression and Metabolic Flux
1.5.3 Gaussian Mixture Model Attractors
1.6 Conclusion
References
2 Functional Roles of Yuragi in Biosystems
2.1 Introduction
2.2 How Muscle Works
2.2.1 Biological Molecular Motor in Muscle
2.2.2 Single Molecule Imaging and Nano-Detection
2.2.3 Bias Brownian Motion Model (Yuragi Model)
2.3 How the Human Brain Recognizes Puzzling Figures by Means of Yuragi Activity
2.3.1 Yuragi Activity in the Human Brain
2.3.2 Psychophysical Experiment of Hidden-Figure Recognition
2.3.3 Yuragi Model of Hidden-Figure Recognition
2.4 Conclusion
References
3 Next-Generation Bio- and Brain-Inspired Networking
3.1 Yuragi-Based Routing and Other Network Control Methods
3.2 Multi-Dimensional Yuragi Model
3.3 Application to Single Network Control
3.3.1 Multipath Routing
3.3.2 Routing in Mobile Ad Hoc Networks
3.4 Application to Multi-Network Control
3.4.1 Multipath Routing in Layered Networks
3.4.2 Cluster-Based Routing in Wireless Sensor Networks
3.4.3 Network Resource Allocation to Multiple Applications on Multiple Vehicles
3.5 Exploration of Better Attractors
3.6 Conclusion
References
4 Yuragi-Based Virtual Network Control
4.1 Introduction
4.2 Attractor Selection
4.2.1 Concept of Attractor Selection
4.2.2 Cell Model
4.2.3 Mathematical Model of Attractor Selection
4.3 Virtual Network Control Based on Attractor Selection
4.3.1 Virtual Network Control
4.3.2 Overview of Virtual Network Control Based on Attractor Selection
4.3.3 Dynamics of Virtual Network Control
4.3.4 Attractor Structure
4.4 Attractor Structure Design
4.4.1 Problem Formulation
4.4.2 Dynamic Reconfiguration of Attractor Structure
4.4.3 Design of Diverse Attractor Structures
4.4.4 Scalable Design of Attractor Structure by Graph Contraction
4.5 Related Work
4.6 Conclusion
References
Part II Yuragi Learning: Extension to Artificial Intelligence
5 Introduction to Yuragi Learning
5.1 Yuragi Learning: An Introduction
5.2 Bayesian Attractor Model for Human Perceptual Decision-Making
5.2.1 Overview
5.2.2 Inference Mechanism for Decision-Making by Bayesian Attractor Model
5.2.3 Design Choices for Bayesian-Attractor-Model-Based Network Control
5.2.3.1 Setting Parameters r and q
5.2.3.2 How to Determine a Criterion for Decision-Making
5.2.3.3 Preparing Attractors
5.3 Virtual Network Reconfiguration Based on Yuragi Learning
5.3.1 Overview of Virtual Network Reconfiguration
5.3.2 Virtual Network Reconfiguration Algorithm
5.3.2.1 Preparation
5.3.2.2 (Step 1) Calculate Confidence Using the BAM-Based Approach
5.3.2.3 (Step 2) Change the Control Phase and Execute the Control
5.4 Performance Evaluation of Yuragi Learning
5.4.1 Evaluation Environments
5.4.2 Characteristics of Virtual Network Reconfiguration Framework
5.4.3 Advantages of Virtual Network Reconfiguration Framework
5.4.4 Impact of the Number of Attractors
5.5 Yuragi Learning with Linear Regression
5.5.1 Virtual Network Reconfiguration Algorithm with Linear Regression
5.5.1.1 (Step 1) Fit the Traffic Situation by Linear Regression
5.5.1.2 (Step 2) Calculate a New VN
5.5.2 Effect of Linear Regression
5.6 Preparing/Updating Attractors in Yuragi Learning
5.6.1 Approach for Preparing Attractors
5.6.2 Approach for Updating Attractors
5.7 Conclusion
References
6 Fast/Slow-Pathway Bayesian Attractor Model for IoT Networks Based on Software-Defined Networking with Virtual Network Slicing
6.1 Introduction
6.2 Bayesian Attractor Model
6.2.1 Decision-Making Process of the Brain
6.2.2 The Analytical Model of BAM
6.2.3 Fast/Slow-Pathway Bayesian Attractor Model
6.3 Proposed Architecture
6.4 Simulation Results
6.5 Conclusion
References
7 Application to IoT Network Control: Predictive Network Control Based on Real-World Information
7.1 Introduction
7.2 Predictive Network Control Based on Yuragi Learning
7.2.1 Model of Human Cognition
7.2.1.1 Abstraction
7.2.1.2 Generative Model
7.2.1.3 Update of State
7.2.1.4 Decision-Making
7.2.2 Application of Yuragi Learning to Predictive Network Control
7.2.2.1 Overview
7.2.2.2 Options and Network Configuration Corresponding to Each Option
7.2.2.3 Abstraction in Predictive Network Control
7.2.2.4 Generative Model in Predictive Network Control
7.2.2.5 Update in Predictive Network Control
7.2.2.6 Decision-Making in Predictive Network Control
7.3 Hierarchical Predictive Network Control Based on Yuragi Learning: Resource Allocation Among Network Slices
7.3.1 Overview
7.3.2 Network Slice Controller
7.3.2.1 Options and Network Configuration Corresponding to Each Option
7.3.2.2 Identification of Current Condition
7.3.2.3 Configuration of Network Slice
7.3.3 Resource Allocation Controller
7.3.3.1 Observations in Resource Allocation Controller
7.3.3.2 Options and Network Configuration Corresponding to Each Option
7.3.3.3 Identification of Current Condition
7.3.3.4 Resource Allocation
7.4 Simple Example
7.4.1 Scenario
7.4.1.1 Network
7.4.1.2 Network Slices
7.4.1.3 Traffic and Real-World Information
7.4.1.4 Controller Settings
7.4.2 Results
7.5 Conclusion
References
8 Another Prediction Method and Application to Low-Power Wide-Area Networks
8.1 Introduction
8.2 Bayesian Attractor Model
8.2.1 Generative Model
8.2.2 State Estimation by Bayesian Filters
8.2.3 Comparison of Bayesian Filters in the Bayesian Attractor Model
8.3 Methods for Channel Congestion Prediction and Channel Assignment
8.4 Evaluation
8.4.1 Simulation Settings
8.4.2 Simulation Results
8.5 Conclusion
References
9 Artificial Intelligence Platform for Yuragi Learning
9.1 Introduction
9.2 Overview of a Brain-Inspired Cognitive Computing System
9.2.1 Conceptual Design of a Brain-Inspired Cognitive Computing System
9.2.2 Architecture Design
9.3 Overview of Yuragi Learning General-Purpose Data Analysis Platform (YGAP)
9.4 Example of YGAP Usage
9.4.1 Preparation of Analysis Data and Configuration of Data Files
9.4.2 Setup of YGAP Console
9.4.3 Training the Model Using Training Data
9.4.4 Classification of Test Data
9.5 Conclusion
References
10 Bias-Free Yuragi Learning
10.1 Introduction
10.2 Classification System with Yuragi Learning
10.2.1 Feature Extractor for Preprocessing of Classification
10.2.2 Yuragi Learning as a Classifier
10.2.3 Yuragi Learning: State Update
10.2.4 Yuragi Learning: Decision-Making
10.3 New Category Acquisition in Yuragi Learning
10.3.1 Detecting New Category
10.3.2 Adding New Category with Initial Data
10.3.3 Gathering Training Data
10.3.4 Updating New Category with Gathered Data
10.4 Numerical Simulation
10.4.1 Simulation Scenario
10.4.2 Accuracy and Sensitivity
10.4.3 Using a Neural Network as Classifier
10.4.4 Results
10.5 Handwritten Character Recognition
10.5.1 Evaluation Scenario with Handwritten Character
10.5.2 Using a Convolutional Neural Network as a Feature Extractor
10.5.3 Using Neural Network as Classifier
10.5.4 Results
10.6 Summary
References
Index
