Table 1:
Comparison of the average unit activations in response to MEIs in two paradigms 1) within architectures and 2) cross architectures, for two architectures ResNet and Attention CNN.
| Paradigm | Within (ResNet) | Cross (ResNet) | Within (Attention) | Cross (Attention) |
|---|---|---|---|---|
|
| ||||
| Gradient Ascent | 11.43 | 5.51 | 7.59 | 4.42 |
| EGG Diffusion | 11.76 | 6.53 † | 10.56 † | 5.50 † |
Bold marks the method which has higher mean activation, and the † marks the increases which are statistically significant (Wilcoxon signed-rank test, respective p-values: 0.08, 2.87·10−6, 2.84·10−10, 4.39·10−5).