**One to many**A method of binary classification is provided. Since there are n feasible solutions to a classification problem, the one to many solution includes N separate binary classifiers, and each possible result corresponds to a binary classifier. During the training, the model will train a series of binary classifiers so that each classifier can answer separate classification questions. Take a picture of a dog as an example. You may need to train five different recognizers, four of which regard the picture as a negative sample (not a dog) and one as a positive sample (dog). Namely:

- Is this a picture of an apple? no
- Is this a picture of a bear? no
- Is this a picture of candy? no
- Is this a picture of a dog? yes
- Is this a picture of an egg? no

When the total number of categories is small, this method is more reasonable, but with the increase of the number of categories, its efficiency will become more and more inefficient.

We can create a significantly more efficient one to many model with the help of a deep neural network in which each output node represents a different category. The following figure shows this approach:

This work adoptsCC agreement, reprint must indicate the author and the link to this article