The Utility of Feature Reuse: Transfer Learning in Data-Starved Regimes

• DOI: 10.48550/arxiv.2003.04117
• Publication Date: 2020-01-01
• Authors: Edward Verenich; Alvaro Velasquez; M. G. Sarwar Murshed; Faraz Hussain
• Keywords: Transfer of learning; Feature (linguistics); Training set; Feature Learning

The use of transfer learning with deep neural networks has increasingly become widespread for deploying well-tested computer vision systems to newer domains, especially those limited datasets. We describe a case domain data-starved regime, having fewer than 100 labeled target samples. evaluate the effectiveness convolutional feature extraction and fine-tuning overparameterized models respect size training data, as well their generalization performance on data covariate shift, or out-of-distribution (OOD) data. Our experiments demonstrate that both overparameterization reuse contribute successful application in image classifiers regimes. provide visual explanations support our findings conclude enhances CNN architectures