GRSEB9S / 1-bit-per-weight

Training wide residual networks for deployment using a single bit for each weight

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Training wide residual networks for deployment using a single bit for each weight

Author: M. D. McDonnell

Contact: mark.mcdonnell@unisa.edu.au

This code was used for experiments accepted for publication in ICLR 2018 (iclr.cc). The ICLR version of the paper and the double-blind open peer review can be found at https://openreview.net/forum?id=rytNfI1AZ (download the PDF here: https://openreview.net/pdf?id=rytNfI1AZ )

The paper is also on arxiv: https://arxiv.org/abs/1802.08530

This page will be populated fully in coming days.

For now, we provide here code written in Matlab and using the matconvnet package (http://www.vlfeat.org/matconvnet/) that will enable the reader to verify our strong error rate results using a single-bit for each weight in inference.

About

Training wide residual networks for deployment using a single bit for each weight

License:MIT License


Languages

Language:MATLAB 100.0%