A Novel Approach Based on Encoder-Decoder with Attention Framework for Multi-modal Multi-label Learn
方法给定实例的输入模态序列{x1^11,x2^22,…,xm^mm}，考虑到每个模态的维度数不同，修改输入模态x~k\tildex^kx~k为[0,0,…,xkx^kxk,…,0]，h→\overrightarrowhh,h←\overleftarrowhh分别是前向和后向隐藏状态，对于第ttt步，h→t\overrightarrowh_tht=Fen→\overrightarrow{F_{en}