关于代码

[suntea233]

c = torch.sum(aux_output * aux_output, dim=-1) gate_value = c.unsqueeze(-1) reverse_gate_value = torch.neg(gate_value).add(1)

c是由两个aux_output点乘，请问这里的c有什么意义呢

[yangli-hub]

hi, this part implements the Dynamic Multimodal Gate that introduced in section 4.4 of the paper.