Table 5.

Bonferroni-corrected p-values of independent t-tests for each pair of transfer learning strategies.

	Only bottleneck	Medium finetuning	Deep finetuning	Finetuning	Training from scratch
Only bottleneck	*	1.67e − 7	2.56e − 4	4.40e − 3	2.23e − 1
Medium finetuning	*	*	2.26e − 1	5.67e − 1	1.30e − 2
Deep fine-tuning	*	*	*	1.0	7.29e − 3
Finetuning	*	*	*	*	8.84e − 3
Training from scratch	*	*	*	*	*

The two background classes were not taken into account in the calculations. All tests were two-sided, with an overall alpha level of α = 0.05. Mean differences where p < 0.05 are highlighted in green.