A Memristive Circuit for Gait Pattern Classification Based on Self-Organized Axon Growth

Abstract

Circuit implementations of neuronal networks should also consider dynamic axon models, since this introduces an additional dynamic aspect due to the transmission delays depending on the axon length. In this work, we derive an electrical circuit for self-organized axon growth based on which we design a neuronal network for learning and classifying gait patterns. We do so by utilizing a wave digital model of the axon model with growth concept, from which we can deduce the corresponding electrical circuit. Here, the axon growth is based on Jaumann structures with memristors. Emulation results show that after the successful training of the network, it can indeed recognize the correct gait patterns. In contrast to typical neuronal networks, this training is not based on synaptic weight changes but on the self-organized axon growth and hence delay-selection. Due to the additional degree of freedom, this can allow for a richer dynamic behavior of modeled neuronal networks.