Artificial intelligence

Abstract: We introduce a new regularization technique column drop which uses inner structure of CNNs for classification to make its output invariant to random crops of input image. Use of this regularization eliminates need in data augmentation by random image cropping under some conditions on architecture of CNN. We show that application of column drop to pooling layers leads to improvement in generalization compared with use of dropout for pooling layers.

Keywords:

References:

1. Goodfellow I, Bengio Y, Courville A (2016)Deep Learning. MIT Press, http://www.deeplearningbook.org
2. Srivastava N, Hinton G, Krizhevsky A, Sutskever I,Salakhutdinov R (2014) Dropout: A simple way toprevent neural networks from overfitting. Journal ofMachine Learning Research 15:1929–1958.
3. A. Krizhevsky, I. Sutskever, and G. E. Hinton.Imagenet classification with deep convolutionalneural networks. In NIPS, 2012.
4. Huang G, Liu Z, van der Maaten L, WeinbergerKQ (2017) Densely connected convolutionalnetworks. In: Proceedings of the IEEEConference on Computer Vision and PatternRecognition
5. Bishop C.M. Pattern Recognition and MachineLearning. – Springer Science + Business Media2006. – 703 p.
6. Krizhevsky A., Learning Multiple Layers offeatures from tiny images, 2009.
7. S. Ioffe and C. Szegedy. Batch normalization:Accelerating deep network training by reducinginternal covariate shift. In ICML, 2015
8. X. Glorot, A. Bordes, and Y. Bengio. Deep sparserectifier neural networks. In AISTATS, 2011.