Deep learning has emerged as the leading technology in various disciplines for its superior performance. Recent research demonstrates that deep learning would be successfully applied to process seismic data. Even though deep neural networks enable superior performance, their lack of intuitive and understandable explanations makes this technique become less trustworthy. In this abstract, we introduce a technique, , to give visual explanations from a convolutional neural network (CNN) that is trained for fault detection. This technique highlights the regions of an input that are particularly influential to the final classification, which is often called the sensitivity map. By analyzing the sensitivity map generated by the CNN trained for fault detection, we find that the CNN does learn some features that are effective for fault detection, but the way a CNN gives its interpretation is still far from human interpretation.
Presentation Date: Tuesday, October 16, 2018
Start Time: 1:50:00 PM
Location: Poster Station 2
Presentation Type: Poster
