Publication: STDP-Based Classificational Spiking Neural Networks Combining Rate and Temporal Coding
Дата
2021
Авторы
Journal Title
Journal ISSN
Volume Title
Издатель
Аннотация
© 2021, The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG.This paper proposes a classification algorithm comprising a multi-layer spiking neural network with Spike-Timing-Dependent Plasticity (STDP) learning. Two layers are trained sequentially. In the rate encoding layer, learning is based on the STDP effect of output spiking rate stabilization. The first layer’s output rates are re-encoded into spike times and presented to the second layer, where learning is based on the effect of memorizing repeating spike patterns. Thus, the first layer acts as a transformer of the input data, and the temporal encoding layer decodes the first layer’s output. The accuracy of the two-layer network is 96% on the Fisher’s Iris dataset and 95% on the Wisconsin breast cancer dataset, which outperforms a sole first layer if the latter involves decoding by rule-interpreting output spike rates. The result resolves the lack of efficient decoding rules for the rate-stabilization-based learning algorithm, and shows the principal possibility of stacking layers with different input encoding and learning algorithms.
Описание
Ключевые слова
Цитирование
STDP-Based Classificational Spiking Neural Networks Combining Rate and Temporal Coding / Serenko, A. [et al.] // Studies in Computational Intelligence. - 2021. - 925 SCI. - P. 403-411. - 10.1007/978-3-030-60577-3_48