Joint Transfer Extreme Learning Machine with Cross-Domain Mean Approximation and Output Weight Alignment

Table 5

Ablation study on different mechanisms of JTELM on USPS + MNIST and Office + Caltech datasets.

CDMA	×	√	×	√	√
OWA	×	×	√	√	√
WA	×	×	×	×	√
USPS vs. MNIST	30.70	45.30	52.55	60.20	62.00
MNIST vs. USPS	51.39	82.56	71.89	75.67	76.39
C⟶A (1)	50.63	53.86	54.80	55.43	57.30
C⟶W (2)	42.71	52.20	54.24	54.58	55.59
C⟶D (3)	42.04	47.13	47.13	48.41	52.86
A⟶C (4)	43.01	42.83	39.57	40.16	41.85
A⟶W (5)	37.63	44.41	48.14	48.81	49.49
A⟶D (6)	42.68	47.13	46.50	47.77	49.68
W⟶C (7)	32.24	34.11	36.24	36.15	36.60
W⟶A (8)	36.85	38.10	39.56	41.34	42.28
W⟶D (9)	80.25	79.62	77.71	80.25	82.17
D⟶C (10)	30.10	33.14	35.00	37.13	37.49
D⟶A (11)	31.52	36.54	41.75	41.96	43.63
D⟶W (12)	71.86	74.58	83.39	85.08	83.73
Total average	44.54	50.82	52.03	53.78	55.08