Skip to main content

Table 4 Correlations of each transferability measure with CNN accuracy for individual algorithms as well as overall

From: Comparison of deep transfer learning algorithms and transferability measures for wearable sleep staging

Measure

Head Re-train, \(n = 144\)

CORAL, \(n = 144\)

Per-Class CORAL, \(n = 144\)

SA, \(n = 144\)

DDC, \(n = 144\)

Overall, \(n = 720\)

LEEP

− 0.03

0.24**

0.28***

0.30***

− 0.11

− 0.07

H-score

0.36***

0.47***

0.47***

− 0.05

− 0.10

0.23***

Hypothesis margin

− 0.07

0.05

0.07

− 0.11

0.15

0.08

Silhouette score

− 0.31

− 0.21*

− 0.21*

0.13

0.07

− 0.11*

MMD, \(\gamma = 0.1\Gamma \)

− 0.25**

− 0.14

− 0.11

− 0.36***

0.33***

− 0.11**

MMD, \(\gamma = 1\Gamma \)

0.07

0.08

0.03

− 0.44***

0.05

− 0.04

MMD, \(\gamma = 10\Gamma \)

0.16

0.29***

0.25**

− 0.53***

0.07

0.05

TDAS, \(\epsilon = 0.1m\)

0.17*

0.24**

0.28***

− 0.17*

0.01

0.11***

TDAS, \(\epsilon = 1m\)

− 0.36***

− 0.42***

− 0.42***

0.19***

0.08

− 0.18***

TDAS, \(\epsilon = 10m\)

− 0.35***

− 0.41***

− 0.42*

0.19*

0.07

− 0.18***

  1. *p < 0.05
  2. **p < 0.01
  3. ***p < 0.001