AAR-CNNs: Auto Adaptive Regularized Convolutional Neural Networks

Authors: Yao Lu, Guangming Lu, Yuanrong Xu, Bob Zhang

IJCAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Comparative experiments are performed on low resolution Image Net, CIFAR and SVHN datasets. Experimental results show that the AAR-CNNs can achieve stateof-the-art performances on these datasets.
Researcher Affiliation Academia 1 Shenzhen Graduate School, Harbin Institute of Technology, China 2 Department of Computer and Information Science,University of Macau, Macau
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide concrete access to source code for the methodology described.
Open Datasets Yes Comparative experiments are performed on low resolution Image Net, CIFAR and SVHN datasets. [...] Low Resolution Image Net datasets [Chrabaszcz et al., 2017] are the downsampled variants of Image Net [Deng et al., 2009] [...] CIFAR [Krizhevsky and Hinton, 2009] datasets include CIFAR-10 and CIFAR-100. They both have 60, 000 32 32pixel colored nature scene images in total, including 50, 000 images for training and 10, 000 images for testing in 10 and 100 classes. [...] The Street View House Numbers (SVHN) dataset [Netzer et al., 2011] includes 73,257 training images, 26,032 testing images and 531,131 images for additional training.
Dataset Splits No For ImageNet, the paper mentions following augmentation details from a prior work but does not specify explicit train/validation/test splits. For CIFAR and SVHN, it specifies train and test set sizes, but no explicit mention of a validation split or its size/methodology is provided.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper mentions utilizing 'SGD method' and 'Nesterov momentum' but does not specify any software libraries or frameworks with version numbers (e.g., PyTorch, TensorFlow versions).
Experiment Setup Yes But the mini-batch size is set to 100. The training epoches are set to 40 and 80 and utilizes the SGD method and Nesterov momentum [Sutskever et al., 2013b] to optimize. The optimization starts from the initial learning rate with 0.01 and 0.1, which is respectively divided by 5 every 10 epoches and divided by 10 at 50% and 75% of the total number of training epochs. The momentum is 0.9, and the weight decay is respectively set to 5e 4 and 1e 4 on AAR-WRN and AAR-DN.