Tensor Switching Networks

Authors: Chuan-Yung Tsai, Andrew M. Saxe, Andrew M. Saxe, David Cox

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experimental results demonstrate that the TS network is indeed more expressive and consistently learns faster than standard Re LU networks.
Researcher Affiliation Academia Chuan-Yung Tsai , Andrew Saxe , David Cox Center for Brain Science, Harvard University, Cambridge, MA 02138 {chuanyungtsai,asaxe,davidcox}@fas.harvard.edu
Pseudocode No The paper describes algorithms textually but does not include any structured pseudocode or algorithm blocks.
Open Source Code Yes The source code and scripts for reproducing our experiments are available at https://github.com/coxlab/tsnet.
Open Datasets Yes We adopt 3 datasets, viz. MNIST, CIFAR10 and SVHN2
Dataset Splits Yes We adopt 3 datasets, viz. MNIST, CIFAR10 and SVHN2, where we reserve the last 5,000 training images for validation.
Hardware Specification No The paper mentions 'multicore CPU acceleration' and 'GPU acceleration' but does not specify any particular hardware models (e.g., CPU or GPU types).
Software Dependencies No The paper mentions software like Matlab, libsvm-compact, Python, Numpy, and Keras, but no specific version numbers for any of these dependencies are provided.
Experiment Setup Yes For all MLPs and CNNs, we universally use SGD with learning rate 10 3, momentum 0.9, L2 weight decay 10 3 and batch size 128 to reduce the grid search complexity by focusing on architectural hyperparameters.