Training Binary Neural Networks using the Bayesian Learning Rule

Authors: Xiangming Meng, Roman Bachmann, Mohammad Emtiyaz Khan

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we present numerical experiments to demonstrate the performance of Bayes Bi NN on both synthetic and real image data for different kinds of neural network architectures.
Researcher Affiliation Academia 1RIKEN Center for Advanced Intelligence Project (AIP), Tokyo, Japan. 2 Ecole polytechnique f ed erale de Lausanne (EPFL), Lausanne, Switzerland.
Pseudocode Yes STE makes a particular choice for the step 1 where a sign function is used to obtain the binary weights from the real-valued weights (see Table 1 for a pseudocode).
Open Source Code Yes The code to reproduce the results is available at https:// github.com/team-approx-bayes/Bayes Bi NN.
Open Datasets Yes We now present results on three benchmark real datasets widely used for image classification: MNIST (Le Cun & Cortes, 2010), CIFAR-10 (Krizhevsky & Hinton, 2009) and CIFAR-100 (Krizhevsky & Hinton, 2009).
Dataset Splits Yes For all the experiments, standard categorical cross-entropy loss is used and we take 10% of the training set for validation and report the best accuracy on the test set corresponding to the highest validation accuracy achieved during training.
Hardware Specification No The paper mentions using the 'RAIDEN computing system' for experiments but does not provide specific details such as exact GPU/CPU models, processor types, or memory amounts.
Software Dependencies No The paper mentions various optimizers and frameworks (e.g., Adam, Bop, Pytorch) but does not provide specific version numbers for any of the software dependencies used in the experiments.
Experiment Setup Yes The details of the experimental setting, including the detailed network architecture and values of all hyper-parameters, are provided in Appendix B.2 in the supplementary material.