The 28th Annual Conference of the Japanese Society for Artificial Intelligence, 2014
The 28th Annual Conference of the Japanese Society for Artificial Intelligence, 2014
AlgorithmSelf-training with confidence-based filtering Input:DL={(x(i), y(i))}L
i=1,DU={x(
i)}N i=L+1,C
Output: f:X → Y
1: Build classifierfLwithDL. 2:D∗←DL
3:FORiinL+ 1 toN
4: ˆy←fL(x(i))
5: IF Conf(ˆy)> C THEN 6: D∗←D∗∪(x(i),yˆ)
7: ENDIF
8: Build classifierf∗withD∗.