Selected article for: "propose approach and transfer learning"
Author: Xuehai He; Xingyi Yang; Shanghang Zhang; Jinyu Zhao; Yichen Zhang; Eric Xing; Pengtao Xie
Title: Sample-Efficient Deep Learning for COVID-19 Diagnosis Based on CT Scans
  • Document date: 2020_4_17
    • ID: l3f469ht_5
    Snippet: Based on these findings, we propose Self-Trans, a selfsupervised transfer learning approach where contrastive selfsupervised learning [3] is integrated into the transfer learning process to adjust the network weights pretrained on source data, so that the bias incurred by source data is reduced. In self-supervised learning (SSL), we construct auxiliary tasks on CT images where the supervised labels in these tasks are solely from the images themse.....
    Document: Based on these findings, we propose Self-Trans, a selfsupervised transfer learning approach where contrastive selfsupervised learning [3] is integrated into the transfer learning process to adjust the network weights pretrained on source data, so that the bias incurred by source data is reduced. In self-supervised learning (SSL), we construct auxiliary tasks on CT images where the supervised labels in these tasks are solely from the images themselves without using any human annotations. Then we adjust the network weights by solving these auxiliary tasks. In these auxiliary tasks, the input images are in the same domain as the data in the target task and no human-annotated labels are used. Therefore, the bias to source images and their class labels can be effectively reduced.

    Search related documents:
    Co phrase search for related documents
    • auxiliary task and target task: 1
    • auxiliary task and transfer learning: 1, 2, 3
    • auxiliary task and transfer learning approach: 1
    • auxiliary task and transfer learning process: 1
    • class label and CT image: 1
    • class label and transfer learning: 1
    • class label and transfer learning approach: 1
    • CT image and image source: 1, 2