EigenNet: Towards Fast and Structural Learning of Deep Neural Networks

Authors: Ping Luo

IJCAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Both the training wall-clock time and number of updates are reduced by using Eigen Net, compared to stochastic gradient descent on various datasets, including MNIST, CIFAR-10, and CIFAR-100.
Researcher Affiliation Academia Ping Luo Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China Department of Information Engineering, The Chinese University of Hong Kong, Hong Kong pluo@ie.cuhk.edu.hk
Pseudocode Yes Algorithm 1 summarizes the training procedure.
Open Source Code No The paper does not provide any explicit statement about releasing its source code or a link to a code repository for the described methodology.
Open Datasets Yes Eigen Net is extensively evaluated on three datasets, including an ablation study on MNIST [Larochelle et al., 2007], a compatibility study with the recent advanced deep architectures, NIN [Lin et al., 2014] and Inception [Ioffe and Szegedy, 2015], on the CIFAR-10 and CIFAR-100 datasets [Krizhevsky, 2009].
Dataset Splits No The paper specifies training and test set sizes for MNIST ('12,000 randomly rotated digit images for training and 50,000 images for test') but does not explicitly mention a separate validation dataset split for any of the experiments.
Hardware Specification No The paper states that 'All models are trained by splitting a batch of 256 samples on 8 GPUs' but does not specify the model or type of GPU, or any other hardware components like CPU or memory.
Software Dependencies No The paper mentions deep learning frameworks and optimizers but does not provide specific version numbers for any software dependencies like Python, PyTorch, or TensorFlow libraries.
Experiment Setup Yes For all algorithms, we initialize network parameters by sampling from a standard normal distribution and employ a batch of 64 samples for each update, where the parameters are updated with a momentum value of 0.9. In particular, for Eigen Net, the whitening interval γ, the number of mini-batches τ, and the pruning threshold ρ are selected from {100,200,500,1000}, {10,20,30}, and {0.01,0.001}, respectively.