Fast Single-Class Classification and the Principle of Logit Separation

Gil Keren, Sivan Sabato, Björn Schuller

We consider neural network training, in applications in which there are many possible classes, but at test-time, the task is a binary classification task of determining whether the…