查看文章

In this chapter, we describe a solution to the problem of applying a deep convolutional survival model to a multimodal scenario and explore a way of training powerful and stable models based on the small samples of multimodal data. We present a new DeepMultiSurv model that was developed to fulfill this goal. The proposed model consists of two parts, feature generators and a high-level association learner. The first part learns the features of single-view data. In the second part, a high-level association between multiview features is learned under the guidance of survival information.We develop two learning strategies to solve the problem of data sets that have small sample sizes.