Multisource Mobile Transfer Learning Algorithm Based on Dynamic Model Compression

Table 3

Pruning rate, accuracy (%), and decrease point of the algorithm on the digit dataset.

Algorithms	U.M- > S			U.S- > M			M.S- > U
	Pruning (%)	Accuracy (%)	Decrease Point	Pruning (%)	Accuracy (%)	Decrease Point	Pruning (%)	Accuracy (%)	Decrease Point
DCTN	0	78.02		0	97.58		0	93.64
	30	77.88	0.14	30	97.47	0.11	30	93.51	0.13
	50	77.7	0.32	50	97.39	0.19	50	93.39	0.25
	70	77.6	0.42	70	97.27	0.31	70	93.18	0.46
	90	76.49	1.53	90	96.14	1.44	90	92.31	1.33
	95	49.56	28.46	95	69.22	28.36	95	64.52	29.12
MultiSource TrAdaBoost	0	78.83		0	96.41		0	93.82
	30	78.72	0.11	30	96.28	0.13	30	93.5	0.32
	50	78.62	0.21	50	96.2	0.21	50	93.44	0.38
	70	78.26	0.57	70	96.15	0.26	70	93.29	0.53
	90	77.4	1.43	90	95.05	1.36	90	92.41	1.41
	95	48.57	30.26	95	66.43	29.98	95	63.71	30.11
M^3SDA	0	79.28		0	99.05		0	95.93
	30	79.21	0.07	30	98.99	0.06	30	95.82	0.11
	50	79.13	0.15	50	98.96	0.09	50	95.7	0.23
	70	79.05	0.23	70	98.89	0.16	70	95.61	0.32
	90	77.72	1.56	90	97.37	1.52	90	94.27	1.66
	95	49.92	29.36	95	69.59	29.46	95	66.95	28.98
MMTLDMC	0	81.23		0	99.13		0	96.95
	X	X	X	X	X	X	X	X	X
	50	81.1	0.13	50	99.07	0.06	50	96.79	0.16
	70	81.07	0.16	70	98.99	0.14	70	96.69	0.26
	90	79.87	1.36	90	97.68	1.45	90	95.49	1.46
	95	52.24	28.99	95	69.92	29.21	95	66.68	30.01