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_44
Snippet: In the equation, m = 0.999 is the momentum coefficient and α is the learning rate of query encoder. Only θ q is updated through back-propagation, whereas θ k maintains a weighted average of the past states. As suggested in Figure 4 , we adopt the self-supervised learning prior to supervised training on the COVID-CT data, as a kind of weight initialization. This mechanism has proved in experiments that it can further improve the model's perform.....
Document: In the equation, m = 0.999 is the momentum coefficient and α is the learning rate of query encoder. Only θ q is updated through back-propagation, whereas θ k maintains a weighted average of the past states. As suggested in Figure 4 , we adopt the self-supervised learning prior to supervised training on the COVID-CT data, as a kind of weight initialization. This mechanism has proved in experiments that it can further improve the model's performance in our CT classification task. The detailed algorithm of Self-Trans is shown in Algorithm 1.
Search related documents:
Co phrase search for related documents, hyperlinks ordered by date