Medical Specialty Classification Based on Semiadversarial Data Augmentation

Table 5

The robustness experiment results of the plain training mode and SemiADA + PI training mode, including accuracy under attack (AUA %) and attack successful rate (ASR %).

Attacks	Plain		SemiADA + PI
	AUA %	ASR%	AUA %	ASR%
None	73.0	—	87.7	—
Textfooler [43]	20.1	72.4	65.7	26.2
Deepwordbug [44]	37.4	48.6	59.6	32.0
Textbugger [45]	40.6	43.0	66.0	24.9
BERT-attack [37]	17.2	76.4	40.9	53.6

The black bold values denote the stronger robustness capability among two modes.