Fix your classifier: the marginal value of training the last weight layerDownload PDF

15 Feb 2018 (modified: 21 Apr 2024)ICLR 2018 Conference Blind SubmissionReaders: Everyone
Abstract: Neural networks are commonly used as models for classification for a wide variety of tasks. Typically, a learned affine transformation is placed at the end of such models, yielding a per-class value used for classification. This classifier can have a vast number of parameters, which grows linearly with the number of possible classes, thus requiring increasingly more resources. In this work we argue that this classifier can be fixed, up to a global scale constant, with little or no loss of accuracy for most tasks, allowing memory and computational benefits. Moreover, we show that by initializing the classifier with a Hadamard matrix we can speed up inference as well. We discuss the implications for current understanding of neural network models.
TL;DR: You can fix the classifier in neural networks without losing accuracy
Code: [![github](/images/github_icon.svg) eladhoffer/fix_your_classifier]( + [![Papers with Code](/images/pwc_icon.svg) 2 community implementations](
Data: [CIFAR-10](, [CIFAR-100](, [ImageNet](, [JFT-300M](, [WikiText-2](
Community Implementations: [![CatalyzeX](/images/catalyzex_icon.svg) 2 code implementations](
13 Replies