Implementation of Human Cognitive Bias on Neural Network and Its Application to Breast Cancer Diagnosis

Abstract

The neural network is one of the most successful machine learning models. However, the neural network often requires large amounts of well-balanced training data to ensure prediction accuracy. Meanwhile, human learners can generalize a new concept from even a small quantity of biased examples, simultaneously enlarging knowledge with an increase in experience. As a possible key factor in the ability to generalize, human beings have cognitive biases that effectively support concept acquisition. In this study, to narrow the gap between human and machine learning, we have implemented human cognitive biases into a neural network in an attempt to imitate human learning to enhance performance. Our model, named loosely symmetric neural network, has shown superior performance in a breast cancer classification task in comparison with other representative machine learning methods.

Keywords

• neural network
• cognitive biases
• imbalanced sample data