Development of JavaScript-based deep learning platform and application to distributed training

Authors: Masatoshi Hidaka, Ken Miura, Tatsuya Harada

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

Reproducibility Variable Result LLM Response
Research Type Experimental In the experiments, we demonstrate their practicality by training VGGNet in a distributed manner using web browsers as the client.
Researcher Affiliation Academia Masatoshi Hidaka, Ken Miura & Tatsuya Harada Department of Information Science and Technology The University of Tokyo 7-3-1, Hongo, Bunkyo-ku, Tokyo, Japan {hidaka,miura,harada}@mi.t.u-tokyo.ac.jp
Pseudocode Yes Figure 1: Example of forward calculation of fully-connected layer using Sushi2
Open Source Code Yes The source code is provided as open-source software1. Download code from https://github.com/mil-tokyo
Open Datasets Yes We evaluated them by training LeNet with MNIST dataset (Le Cun et al., 1998b).
Dataset Splits No While the paper mentions using the MNIST dataset and shows 'mnist train' and 'mnist test' in a configuration file example (Figure 2), it does not explicitly provide details about a specific validation dataset split.
Hardware Specification Yes Table 2: Hardware used for the experiments. GPU: AMD FirePro S9170, NVIDIA K80. CPU: Intel Core i7-5930K, Intel Xeon E5-2690 v3.
Software Dependencies Yes Firefox (version 32) and node.js (version 4.3.0) are used as the JavaScript execution environment.
Experiment Setup Yes The network structure is based on Le Cun et al. (1998a), which contains two convolutional layers and two fully-connected layers. The batch size is 64. ... The optimization method is momentum SGD. ... The batch size is 256 according to (Simonyan & Zisserman, 2014a).