TABLE 2. Softmax Results for the Semi-Supervised and Supervised Models With Different Numbers of Labels . Higher Values Indicate Higher Model Confidence. The Higher the Better for Correct Estimations, and the Lower the Better for Incorrect Estimations.

	Desc. stat.	correct NO-SSDL	wrong NO-SSDL	JS div. No SSDL	correct SSDL	wrong SSDL	JS div. No SSDL
20		0.8216	0.7343	0.2407	0.9452	0.7481	0.5812
	s	0.1384	0.1457		0.0932	0.1298
30		0.8594	0.7207	0.3554	0.9597	0.7722	0.5497
	s	0.1337	0.1429		0.0755	0.1556
60		0.9159	0.7128	0.506587	0.9301	0.7315	0.5462
	s	0.1116	0.1504		0.0986	0.1384
70		0.9097	0.7387	0.4514	0.9232	0.7309	0.4976
	s	0.1167	0.1535		0.1066	0.1467
100		0.9324	0.7297	0.5067	0.8445	0.7254	0.2872
	s	0.1016	0.1471		0.1438	0.1505