Is there such a thing as a double sided neural networks? I am trying to see if there are already established algorithms in Neural Networks for matching purposes. Lets assume there are two different parties available each with a separate dataset (like men and women in a dating site). Can we start a Neural work on one side (men) to get an output and conduct another neural Network from the other side (women) which could be possibly totally different and find a way to compare (measure the distance or the similarity) of the two outcomes? Assuming to have a large enough set of data would it possible to train both networks simultaneously? I have looked into Siamese Neural networks for instance, but that would not be the answer since in the case of Siamese networks both branches of the network have to be identical.
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
1
So you have two dataset and the goal is to tell if they match or not and the features of them are different right ? here we have binary classification and its normal ML problem, you can use a neural network ass well for extract features, if the features are so much, then you can classify them final layer with sigmoid function or any ML algorithm such as SVM
-
$\begingroup$ I am sorry Mustafa. I don't quite understand. This is not a binary classification. It is about matching apples to oranges (Men to Women) or People to Places and so on. As another example lets say you want to decide which type of jeans matches to which type of men. There are 50 types of jeans and go and figure how many type of men we got out there. This is a problem that goes towards ad-optimization I would say but not completely. $\endgroup$ Mar 6, 2021 at 16:27