This work is based on a diploma thesis by Changxing Dong [Dong04]. In [Dong04], the effects of different shapespaces on the behaviour of artificial immune system algorithms are evaluated. For that purpose a pattern recognition task was used. In this case, chinese characters should be learned and recognized. Each character was adapted individually using theCLONALGalgorithm [dCT02]. The resulting set of antibody patterns, the duration of running the algorithm and the quality of recognition were then used to evaluate the differences between the different shapespaces.
Table of Contents
1 Introduction
1.1 Background
1.2 Fundamental Ideas
2 The Core Immune Learning Algorithm
2.1 Problem Description
2.2 Deciding the Immune Principles to be used for Problem Solving
2.3 Engineering the Artificial Immune System
2.3.1 Defining the Types of Immune Components to be used
2.3.2 Defining the Mathematical Representation for the Elements of the AIS
2.3.3 Applying Immune Principles to Problem Solving
2.3.4 The Metadynamics of the AIS
2.4 Reverse Mapping from AIS to the Real Problem
2.5 The Full Learning Algorithm in Pseudocode
2.5.1 The aiNet Algorithm
2.5.2 The Isis Algorithm
2.5.3 Comparison of aiNet and Isis
3 The Architecture
3.1 The Concept
3.2 The Ra Algorithms
3.2.1 The Learning Algorithm: raCreate
3.2.2 The Classification Algorithm: raCluster
4 Implementation
4.1 Implemented Classes
4.2 Invoking the Algorithms
4.2.1 Isis
4.2.2 raCreate
4.2.3 raCluster
4.2.4 raGraph
5 Data Preprocessing
6 Results
6.1 Results with the Identity Shapespace
6.1.1 Level 4 Clusters with Identity Shapespace
6.1.2 Comments
6.2 Results with the Convolution Shapespace
6.2.1 Level 2 Clusters with Convolution Shapespace
6.2.2 Comments
6.3 Results with the Fourier Shapespace
6.3.1 Level 1 Clusters with Fourier Shapespace
6.3.2 Comments
6.4 Comparison of the Algorithm Behaviour Across the Different Shapespaces
7 Further Possibilities for the Ra/Isis Approach
7.1 Parametertuning
7.1.1 Hierarchical Parameters
7.2 Shapespaces
7.2.1 Hierarchical Shapespaces
7.3 Clustering Unusual Data
7.4 Data Analysis and Other Learning Algorithms
7.5 Classification
7.5.1 Hierarchical Classification
7.5.2 Classification Networks
7.5.3 Subsequent Classification
8 Analysis of the Algorithm Behaviour
8.1 Ra/Isis vs. Common Aglomerative Clustering
8.2 Comparison of Ra/Isis with Self Organizing Maps
8.2.1 Results of Applying the Data to a SOM
9 Conclusion
Research Objectives and Themes
This work aims to develop an artificial immune system (AIS) algorithm capable of hierarchical clustering of complex input data, specifically Chinese characters represented as binary pixel matrices. By implementing an iterative, recurrent architectural approach called Ra/Isis, the research seeks to overcome the limitations of standard algorithms by grouping patterns based on conceptual similarities, effectively creating a conceptual hierarchy from unlabeled data.
- Development of the Isis clustering algorithm based on immune network theory.
- Implementation of the Ra/Isis recurrent architecture for hierarchical clustering.
- Evaluation of different shapespace transformations (Identity, Convolution, Fourier).
- Comparative analysis of algorithm behavior and clustering performance.
- Exploration of analogies between artificial immune systems and self-organizing maps.
Excerpt from the Book
The Core Immune Learning Algorithm
To design the algorithm, I followed the Guidelines for the Design of AIS from the book Artificial Immune Systems: A New Computational Intelligence Approach [dCT02]. The algorithm developed in this section, is the core clustering algorithm. The usage of this algorithm to create the hierarchy is explained in a later section of this document.
The task to be solved is, as already outlined, to create a system capable of grouping together similiar input patterns. The base algorithm has only to be able to map one or more input patterns to one recognizing pattern. The input patterns are binary pixel matrices, with a dimension of 16 for both x and y directions.
Two main elements can be indentified: the patterns to be mapped and the patterns onto which the input patterns are mapped.
The algorithm is called Isis, an abreviation of Iindependent Single Immune System.
Summary of Chapters
Introduction: Outlines the background of artificial immune system algorithms for pattern recognition and sets the fundamental goal of using a recurrent architecture for hierarchical clustering.
The Core Immune Learning Algorithm: Details the development of the Isis algorithm, focusing on immune network principles, data representation, and subprocesses like clonal selection and network suppression.
The Architecture: Explains the conceptual Ra/Isis setup, which wraps the core Isis algorithm in a recurrent structure to enable hierarchical clustering and data categorization.
Implementation: Describes the Java-based software implementation, including the primary classes like Isis, Matrix, and Ra, and how these components are invoked.
Data Preprocessing: Discusses the transformation of the input character data to optimize the algorithm's capability to identify and group patterns effectively.
Results: Presents the findings from clustering experiments using Identity, Convolution, and Fourier shapespaces, providing visual graphs and detailed cluster analyses.
Further Possibilities for the Ra/Isis Approach: Offers visions for future research, including parameter tuning, iterative shapespace modifications, and applying the framework to unusual data types or classification tasks.
Analysis of the Algorithm Behaviour: Compares the performance and methodology of Ra/Isis against traditional aglomerative clustering and self-organizing maps to better understand its mechanics.
Conclusion: Summarizes the potential of the Ra/Isis framework and highlights the value of comparing artificial immune systems with other unsupervised learning models.
Keywords
Artificial Immune Systems, AIS, Isis Algorithm, Ra/Isis, Hierarchical Clustering, Pattern Recognition, Machine Learning, Shapespace Transformation, Unsupervised Learning, Clonal Selection, Immune Network Theory, Convolution Shapespace, Fourier Shapespace, Self-Organizing Maps, Data Mining.
Frequently Asked Questions
What is the primary purpose of this research?
The research focuses on designing an artificial immune system (AIS) that performs unsupervised hierarchical clustering on complex datasets, specifically Chinese character patterns.
What are the core thematic areas covered?
The work covers algorithm design based on immunology, software architecture for iterative clustering, data preprocessing techniques, and comparative analysis of shapespace transformations.
What is the main research objective?
The goal is to move beyond simple mapping and create a bottom-up hierarchical structure that extracts meaningful clusters from raw input data using the Ra/Isis recurrent architecture.
Which scientific method is utilized for clustering?
The work employs an artificial immune system (AIS) strategy inspired by discrete models of the immune network theory, specifically adapting aiNet concepts to create clusters that represent conceptual features of the input data.
What does the main body of the work address?
It addresses the development of the Isis algorithm, the architectural implementation of the Ra framework, detailed results of clustering experiments with various mathematical transformations, and extensive comparisons with other machine learning methods like SOMs.
Which keywords best characterize this work?
Key terms include Artificial Immune Systems, Hierarchical Clustering, Pattern Recognition, Recurrent Architecture, and Shapespace Transformation.
How do shapespaces influence the clustering outcome?
Shapespaces transform the input raw pixel data into a more abstract representation. Different transformations, such as Fourier or Convolution, highlight specific feature sets, which dramatically changes the generalization performance and cluster formation characteristics.
Why are comparisons with self-organizing maps (SOMs) included?
The author includes these comparisons to demonstrate parallels between the way Ra/Isis adapts to data topology and how SOMs create U-Matrices, suggesting that both methods share underlying mechanisms for structure extraction in high-dimensional data.
- Quote paper
- Stefan Schadwinkel (Author), 2005, A Recurrent Architecture of Independent Single Immune Systems, Munich, GRIN Verlag, https://www.grin.com/document/48178