Skip to content

Latest commit

 

History

History
21 lines (11 loc) · 912 Bytes

README.md

File metadata and controls

21 lines (11 loc) · 912 Bytes

OnDeviceTraining

This is the src of the android project we used to measure the on-device training time of pre-defined DNN.

Note that this project is developed based on the DL4JIrisClassifierDemo, you may need to refer to http://deeplearning4j.org/ for more information.

How to run

The project is built using Android Studio, so the simplest way to run it is importing it to Android Studio and re-building it.

Guide

  • modify the function in onClick to measure the on-device trianing time of pre-defined DNN, e.g. olaf_reddit();
  • modify the hyper-parameter if needed
  • note that we only measure the training time of one batch, so the optimizer, params, initializer do not matter
  • functions we used are olaf_*
  • run initialize(String) first to avoid cold boot