Semi-Supervised Spiking Neural Network for One-Shot Object Appearance Learning

Igor Peric, Robert Hangu, Jacques Kaiser, Stefan Ulbrich, Arne Roennau, Johann Marius Zoellner, Ruediger Dillman

2017

Abstract

We present a network of spiking neurons which extracts intermediate-level features from a set of classes in an unsupervised manner and uses them to later learn new, unrelated classes with just a few training examples, also called one-shot learning. The framework is built on the biologically plausible neurosimulator NEST developed and used by neuroscientists, giving the work an unprecedented biological plausibility over previous similar approaches which use custom-built systems tuned to their needs. Furthermore, the learning of the classes happens in a continuous manner, without scripted interruptions and external interventions to the neuron states during simulation, which draws this work even closer to biological realism. The high quality of the learned features is confirmed by achieving a close to state-of-the-art F1 score of 97% during the recognition of the same classes, while obtaining a score as high as 72% for one-shot learning. This paper focuses more on the biological plausibility of the presented ideas and less on the concrete object classification mechanisms.

Paper Citation

in Harvard Style

Peric I., Hangu R., Kaiser J., Ulbrich S., Roennau A., Zoellner J. and Dillman R. (2017). Semi-Supervised Spiking Neural Network for One-Shot Object Appearance Learning.In Proceedings of the 5th International Congress on Neurotechnology, Electronics and Informatics - Volume 1: NEUROTECHNIX, ISBN 978-989-758-270-7, pages 47-53. DOI: 10.5220/0006503300470053

in Bibtex Style

@conference{neurotechnix17,
author={Igor Peric and Robert Hangu and Jacques Kaiser and Stefan Ulbrich and Arne Roennau and Johann Marius Zoellner and Ruediger Dillman},
title={Semi-Supervised Spiking Neural Network for One-Shot Object Appearance Learning},
booktitle={Proceedings of the 5th International Congress on Neurotechnology, Electronics and Informatics - Volume 1: NEUROTECHNIX,},
year={2017},
pages={47-53},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0006503300470053},
isbn={978-989-758-270-7},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 5th International Congress on Neurotechnology, Electronics and Informatics - Volume 1: NEUROTECHNIX,
TI - Semi-Supervised Spiking Neural Network for One-Shot Object Appearance Learning
SN - 978-989-758-270-7
AU - Peric I.
AU - Hangu R.
AU - Kaiser J.
AU - Ulbrich S.
AU - Roennau A.
AU - Zoellner J.
AU - Dillman R.
PY - 2017
SP - 47
EP - 53
DO - 10.5220/0006503300470053