An Analogy between the use of Back Propagation and Associative Memory as a Model for the Artificial Immune Network Memory

the artificial immune system applications vary from anomaly detection, fault tolerance, as well as data classification and data clustering. It was noticed that the applications on the design of the artificial immune memory are sparse, despite its importance in the learning process within the artificial immune networks. Most of the work presented focused only on the secondary immune response. In this research, the focus is on the artificial immune memory with its main components: the primary response and the secondary response. A novel model is proposed to model the artificial primary immune response using an idiotopic artificial immune network. On the other hand, the artificial secondary immune response is modeled using two techniques: the hetero-associative networks and the back-propagation networks. While modeling the secondary immune response, it was noticed that data representation has an effect on the performance of the network performance. As well, when the secondary immune response was modeled using the back-propagation network, data preprocessing was needed to guarantee error minimization. After applying these necessary data treatments, the backpropagation network outperformed the hetero-associative networks, when noise was presented in the input patterns. For example, after adding 25% noise to the input data, the memory model based on back-propagation networks showed low rate 6.25%, compared to 12.5% for the hetero-associative based model. As well, after adding 50% noise to the input data, the back-propagation based networks showed low rate 18.75%, compared to 31.25% for the hetero-associative based model. Based on previous findings, it is believed that the designed primary and secondary memory models achieve significant potential in representing the behavior of the original immune system memory. In particular, using the idiotpic immune network to represent the primary immune response, and using the back-propagation networks to represent the secondary immune response.