Demystifying Dropout

Authors: Hongchang Gao, Jian Pei, Heng Huang

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results have verified the effectiveness of our proposed method.
Researcher Affiliation Collaboration 1JD Finance America Corporation 2Department of Electrical and Computer Engineering, University of Pittsburgh, Pittsburgh, USA 3JD.com 4School of Computing Science, Simon Fraser University, Canada.
Pseudocode Yes Algorithm 1 Augmented Dropout
Open Source Code No The paper provides links to Caffe library implementations (e.g., 'https://github.com/BVLC/caffe/blob/master/examples/mnist/ lenet_train_test.prototxt') which are third-party tools used by the authors, not their own source code for the methodology described in the paper.
Open Datasets Yes Throughout our experiments, we employ four widely used datasets: MNIST (Le Cun et al., 1998), SVHN (Netzer et al., 2011), CIFAR10 (Krizhevsky, 2009), and CIFAR100 (Krizhevsky, 2009).
Dataset Splits No The paper describes training and testing set sizes (e.g., 'MNIST contains 70,000 handwritten digits images of size 28x28. 60,000 images are used for the training set and the rest 10,000 images are used for the testing set.'), but does not explicitly state validation dataset splits or methodology.
Hardware Specification No No specific hardware details (like exact GPU/CPU models, processor types, or memory amounts) used for running experiments are provided in the paper. Only general training settings are described.
Software Dependencies No The paper mentions using 'Caffe (Jia et al., 2014)' but does not provide a specific version number for Caffe or any other software dependencies.
Experiment Setup Yes The batch size is set to 100. We employ SGD with the momentum of 0.9 as the optimizer. The training procedure starts with a learning rate of 0.001 which is divided by 10 at 4,000 iterations. We terminate the training procedure at 9,000 iterations.