Can I read Compensating Feature Deviation in ConvNets on EtoBox?

Compensating Feature Deviation in ConvNets by 2000010605 is a document available to read on EtoBox.

What is Compensating Feature Deviation in ConvNets about?

1) The document investigates why convolutional neural networks (ConvNets) trained on class-imbalanced data perform poorly on minor classes, unlike traditional machine learning algorithms which often underfit minor classes. 2) It was discovered that ConvNets significantly overfit minor classes due to a phenomenon of "feature deviation" where the features learned for minor classes differ between training and test data. 3) To address this, the authors propose incorporating class-dependent temperatures during

Author

2000010605

Language

EN