Adaptive methods for accelerating Deep Neural Networks on ARM GPU architectures
This repository contains all the codes necessary to reproduce the experiments and the results obtained.
The work has been enclosed in various folders with a name that describes the phase to be addressed.
I recommend following the steps from the first to the last.

Phase1 explains how to create a dataset for experiments.

Phase2 explains how to perform the experiments and collect the results.

Phase3 explains how to read the results and get a .arff file

Phase4 explains how to get a model to use with machine learning techniques

Phase5 explains how to get the results of the analysis and contains some experiments carried out by me

In these example images you can see the bar chart.
The height of the bars indicates the running time.
Please note that at the bottom of each bar is the number of successful experiments on the dataset taken into consideration.
The predicted column indicates in brackets the number of experiments used for each technique: for example (3,7,9) it would indicate 3 experiments obtained with Conv, 7 with Directconv, 9 with Winograd.