Recently I've been working on some deep neural networks for binary classification, and in developing the appropriate features, I found the necessity to classify the input data. For example, if the input feature is consisted of information five different objects listed sequentially, I want to add the "class" each object, increasing the feature size by five.
The problem is that I'm not quite sure how the inputs would be classified, or would be optimally classified. Therefore I was wondering if there was a method to train a network to find the appropriate classification method for the inputs, in order to maximize the training result. I found that the method used for word2vec wouldn't work here because in a sense, this would be unsupervised learning, because there wouldn't be any label for the training.