Analysis on the Dropout Effect in Convolutional Neural Networks

217 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Improved Regularization of Convolutional Neural Networks with Cutout (2017, methodology): “This approach was similar to maxdrop [13], in that we aimed to remove maximally activated features in order to encourage the network to consider less prominent features.”
Deep learning classifiers for hyperspectral imaging: A review (2019, background): “However, traditional dropout injects random single-pixel noise to the feature maps, resulting in spatially unstructured noise, which makes it ineffective in 2-D and 3-D models (Park and Kwak, 2017).”
Attention-Based Dropout Layer for Weakly Supervised Object Localization (2019, methodology): “The drop mask is a dropout-based technique and its rationale is similar to MaxDrop [27].”
See Better Before Looking Closer: Weakly Supervised Data Augmentation Network for Fine-Grained Visual Classification (2019, methodology): “Before our work, random spatial image augmentation methods [19, 3, 24], such as image cropping and image dropping, have been proposed and proved to be effective in improving the robustness of deep models.”
Accurate localization of cortical and subcortical sources of M/EEG signals by a convolutional neural network with a realistic head conductivity model: Validation with M/EEG simulation, evoked potentials, and invasive recordings (2024, background): “The generalizability can be attributed to batch normalization (Ioffe and Szegedy, 2015) and dropout layers (Park and Kwak, 2017).”