"Neural collapse with cross-entropy loss and positive features"Nguyen, Duc AnhNeural Collapse consists of several emergent phenomena which have been empirically observed across various canonical classification problems during the terminal phase of training. Neural Collapse is often studied through a simplified model which has arbitrary unconstrained features. Here, we augment this model with non-negativity of the features, which is motivated by the non-negativity of common activation functions such as ReLU and sigmoid. |
| http://univie.ac.at/projektservice-mathematik/e/talks/Nguyen_2021-06_NC-ICCHA21.pdf |
« back