Domain Adaption Based on ELM Autoencoder

Table 1

(a) Recognition accuracy with DA using a NN classifier (Office dataset + Caltech-256)

Algorithm
NA [7]	21.53%	28.71%	23.64%	22.55%	26.26%	19.21%
DA-SA1 [7]	38.00%	29.80%	35.50%	30.90%	29.60%	31.30%
DA-SA2 [7]	40.50%	33.00%	38.00%	33.30%	31.20%	31.90%
PCA [7]	39.00%	38.00%	37.40%	35.30%	32.40%	32.30%
GFK [16]	36.90%	32.52%	31.10%	35.61%	29.86%	27.26%
SA-ELM-DA(SURF, sigmoid)	39.72%	36.89%	34.42%	35.74%	33.24%	33.61%
SA-ELM-DA(SURF, linear)	25.62%	30.02%	26.31%	24.00%	26.02%	19.00%
SA-ELM-DA(CNN, linear)	20.15%	15.52%	20.92%	17.43%	11.75%	13.47%
SA-ELM-DA(CNN, sigmoid)	58.36%	43.79%	71.83%	49.74%	34.54%	41.63%

(b) Recognition accuracy with DA using a NN classifier (Office dataset + Caltech-256)

Algorithm
NA [7]	21.34%	21.08%	54.01%	25.24%	20.32%	62.44%
DA-SA1 [7]	34.60%	37.40%	71.80%	35.10%	33.50%	74.00%
DA-SA2 [7]	34.70%	36.40%	72.90%	36.80%	34.40%	78.40%
PCA [7]	37.60%	39.60%	80.30%	38.60%	36.80%	83.60%
GFK [16]	35.24%	35.21%	70.63%	34.46%	33.71%	74.92%
SA-ELM-DA(SURF, sigmoid)	36.86%	39.92%	80.83%	36.76%	36.92%	84.41%
SA-ELM-DA(SURF, linear)	30.00%	32.60%	83.38%	30.00%	26.00%	79.47%
SA-ELM-DA(CNN, linear)	18.74%	13.54%	15.76%	17.16%	12.18%	13.64%
SA-ELM-DA(CNN, sigmoid)	46.94%	35.86%	53.82%	52.61%	46.69%	54.69%