Reading: Cutout — Improved Regularization of Convolutional Neural Networks (Image Classification)

Regularization on Input, Improve the Accuracy of ResNet, WRN, and Shake-Shake.

Cutout applied to images from the CIFAR-10 dataset.

Outline

1. Motivation

2. Differences from Dropout

3. Experimental Results

3.1. CIFAR10, CIFAR100, SHVN

Cutout patch length with respect to validation accuracy with 95% confidence intervals (average of five runs).
Test error rates (%) on CIFAR (C10, C100) and SVHN datasets

3.2. STL-10

Test error rates on STL-10 dataset. “+” indicates standard data augmentation (mirror + crop). Results averaged over five runs on full training set.

3.3. Analysis of Cutout’s Effect on Activations

Magnitude of feature activations, sorted by descending value, and averaged over all test samples

PhD, Researcher. I share what I've learnt and done. :) My LinkedIn: https://www.linkedin.com/in/sh-tsang/, My Paper Reading List: https://bit.ly/33TDhxG