Program Results

Introduction to the event

In the era of big data, how to make more accurate predictions is the focus of everyone's attention. Furthermore, the choice of variables is a significant factor that affects predictions. In June 2019, Yu-Shan Scholar, Professor Shaw-Hwa Lo, shared the influential score method, as known as I-Score method, proposed by his research team with Chiao Tung University. I-Score method can select the important variables with considerable influence on prediction from a large number of variables. the proposed method not only considers the interaction between the variables, but also estimates the lower limit of the accuracy of prediction without overfitting. with the sharing by Professor Lo, we expected to utilize this method on medical images. After plenty of video conference discussions during the period, we went to Columbia University in New York in October 2019 to collaborate with Professor Lo’s team, applied I-Score method to the feature selection of deep learning namely CNN classification models. It is observed that I-Score method can effectively figure out a few feature values with interaction and prediction capabilities, while maintaining considerable prediction capabilities. Hence, we look forward to publishing this new study internationally to imporve the development of  deep learning models visualization.