For pseudo - Siamese network loss should be used?

CodePudding user response:

Or you can use the same loss, similarity measure, because we are hope to will all kinds of information through neural network to map to the same space (that is, your so-called coordinate system), it is reasonable, many models are doing
Some recommendation system, for example, "users" and "goods", this seems to be completely different kinds of input to the map to the same low dimensional space, under the same space can be distance measure, used to judge whether the users like or hate a product