Embedded Machine Learning on a Programmable Neuromorphic Platform

Sugiarto, Indar and Gumelar, Agustinus Bimo and Yogatama, Astri (2022) Embedded Machine Learning on a Programmable Neuromorphic Platform. In: 3rd International Conference on Computer Science, Electrical & Electronic Engineering (ICCEE 2021), August 2021, 11-03-2022 - 11-03-2022, Jakarta - Indonesia.

	PDF Download (763Kb)
	PDF Download (3670Kb)

Abstract

This paper presents an implementation of k nearest neighbor (k-NN) algorithm on SpiNNaker. SpiNNaker is a programmable neuromorphic platform well-known for its very low power energy consumption which is suitable to be used as an embedded system. By utilizing the SpiNNaker communication protocols, we are able to efficiently distribute jobs across SpiNNaker cores for performing the k-NN algorithm. From the experiments, we observed that the k-NN program runs smoothly and performs basic classification task on Irish dataset correctly. From the investigation, we found that the k-NN program reports not only the accuracy and the best k-value to get that accuracy, but also the reason why that k-value should be chosen for the corresponding classification task. We also found that increasing the potential number for k-values, the k-NN program was able to find better accuracy. For example, the accuracy of 91.67% was achieved when we increased the range of k between 30 to 60.

Actions (login required)