Skip to main content

Table 6 Different pooling methods for top 10 terms in all stages

From: Image-level and group-level models for Drosophilagene expression pattern annotation

Stages Max pooling
  Accuracy AUC Sensitivity Specificity
4-6 75.43 0.7320 0.7002 0.7637
7-8 77.28 0.7616 0.7311 0.7921
9-10 74.46 0.7457 0.7397 0.7517
11-12 78.51 0.7893 0.8015 0.7771
13-17 80.96 0.7952 0.7664 0.8240
Stages Average pooling
  Accuracy AUC Sensitivity Specificity
4-6 75.21 0.7315 0.6976 0.7654
7-8 76.80 0.7603 0.7332 0.7874
9-10 74.32 0.7439 0.7397 0.7480
11-12 77.49 0.7828 0.8034 0.7623
13-17 80.12 0.7842 0.7521 0.8164
Stages Sqrt pooling
  Accuracy AUC Sensitivity Specificity
4-6 75.73 0.7360 0.7031 0.7692
7-8 77.49 0.7664 0.7373 0.7955
9-10 74.58 0.7459 0.7380 0.7538
11-12 78.37 0.7907 0.8094 0.7720
13-17 80.73 0.7936 0.7681 0.8191
  1. In our experiment, all the three pooling methods produce comparable performance. In particular, max pooling and Sqrt pooling perform slightly better than average pooling.