Artificial Alzheimer's - visualizing the process of memory decay, confusion and death in a simple Neural Network Model of Associative Memory

Abstract

This paper analyzes the behaviour of Alzheimers disease simulation in artificial neural networks and thereby suggests a new possible diagnosis for Alzheimer. Alzheimers disease is one of the most common diseases, increasing severity over time. Despite its high prevalence and thousands of yearly publications in this area, no cure has been found to date and early detection is one of the most important factor towards fighting the disease and possibly finding a cure. This paper analyses the behaviour of simulated Alzheimer in Hopfield memories and observes that the main factor influenced by the loss of connections is the time needed to recognize distorted symbols - while the recognition performance stays surprisingly high for a long time. Hence, I suggest a new early diagnosis approach which is based on slightly distorted elements (e.g., characters) and measures the time needed to perform the task, instead of just focusing on the subjects recognition performance. This insight could have a large impact to future diagnose setups, once being validated in a clinical study.