Natural immune systems for developing novel Machine Learning algorithms

Abstract

This thesis presents novel multi-layered natural immune system (NIS) inspired algorithms in the domain of machine learning. The exploration of biological metaphors for developing novel learning algorithms for computation is not new. The contribution of artificial neural networks inspired by the human brain, genetic algorithms inspired by modern-day genetics, and ant colony algorithms inspired by swarm intelligence is well recognized. A relatively new addition is the artificial immune system (AIS) that is inspired by an NIS, where micro-level processes and metaphors of NIS are used to develop novel computing algorithms.

In recent years, two main AIS algorithms, namely, CLONALG and aiNet, have been developed based on the principles of ‘clonal selection’ and ‘immune network theory’ respectively. Both of these algorithms can be regarded as data reduction processes with limited learning capability. The role of evolutionary computing in these algorithms is restricted to the hypermutation of antibodies in response to pathogens, however. In this thesis, the role of evolutionary computing in AIS is broadened significantly to include the evolution of learning parameters and an active role of memory cells, as well as a population-based approach to AIS. A novel AIS algorithm inspired by the humoral mediated immune response (HIR) triggered by the adaptive immune system is presented. In humoral immune response, antibodies with hypermutation are secreted to mount an appropriate immune response to pathogens. This thesis introduces the concept of a layered architecture, where pathogens are identified and captured using different layers of cells such as antibodies, memory cells, and B-cells.

To date, the focus of AIS researchers has been on developing novel AIS algorithms using various NIS metaphors and processes. Relatively little research has focused on integrating AIS models with other nature-inspired techniques to achieve effective and improved learning capabilities. In this thesis, we propose a novel methodology inspired by the vaccination process in an NIS, where information contained in the memory of an AIS is used to prime learning in a hybrid architecture. We demonstrate the effectiveness of this hybrid architecture and explore future directions of AIS.