A1510
Title: An automatic approach to explainable AI with applications to medical image classification
Authors: Tso-Jung Yen - Academia Sinica (Taiwan) [presenting]
Abstract: Recent advances in explainable AI have inspired researchers to develop methods that can effectively detect input impact on the output of a predictive model. However, most of these methods require multiple stages to process the input. Some of the stages involve re-sampling the input data. When there are large amounts of input data, such a re-sampling procedure may take time to proceed. A method is proposed that can automatically detect input impact on the output of the predictive model. The method relies on the idea that turns the impact detection problem to a prediction problem. Simulation experiments show that the method can correctly carry out the input impact detection without spending too much time on input processing.
