Sliding threshold of spike‐rate dependent plasticity of a semiconducting polymer/electrolyte cell

Abstract

Spike‐rate dependent plasticity, one of the conventional learning protocols in neuroscience, has been achieved in semiconducting polymer/electrolyte cells. The frequency threshold θm of spike‐rate dependent plasticity is sliding in requirement of stability. In this work, various prior signal inputs are applied to poly[2‐methoxy‐5‐(2‐ethylhexyloxy)−1,4‐phenylenevinylene]/polyethylene oxide ‐Nd3+ cells to explore their effects on θm.The study find that a prior inhibitory input, i.e., a weak stimulation, moves θm in the low frequency direction, while a prior excitatory input, i.e., an intensive stimulation, moves θm in the high frequency direction. Our results demonstrate the adaptivity of ionic kinetics in macromolecules systems for signal handling and learning. We suggest that a sliding threshold is due to the variable active ionic domains depending on stimulation strength and history. © 2016 Wiley Periodicals, Inc. J. Polym. Sci., Part B: Polym. Phys. 2016, 54, 2412–2417

An memristor-based synapse implementation using BCM learning rule

“…Particularly, the Hebbian learning rule is a widely used neural network learning mechanism [10]. However, the conventional Hebbian learning rule has the drawback of unlimited modulation of synaptic weight, which can cause the system to collapse [10,11]. Therefore, this learning rule was modified by introducing a sliding threshold, i.e.…”

“…Therefore, this learning rule was modified by introducing a sliding threshold, i.e. the Bienenstock-Cooper-Munro (BCM) learning rule, which can significantly improve the network stability [11,12,13]. Inspired by this, in our previous work [14], a self-repairing learning rule for spiking astrocyte-neuron networks is proposed.…”

“…However, due to the physical properties of the memristor, i.e. the synaptic weight changes when it is stimulated under signals from the pre-or postsynaptic neurons, causing the memristor-based synapse faces the prominent problem of weight infinite modulation [11,20,21]. Considering that, if the synaptic weight of memristor-based synapse is modulated too high (i.e.…”

“…Considering that, if the synaptic weight of memristor-based synapse is modulated too high (i.e. memristance is very low), it leads to an overstimulated network where the postsynaptic neurons collapse [11]. Conversely, if the weight is too low (i.e.…”

“…Conversely, if the weight is too low (i.e. memristance is very high), the memristor-based synapse is in an inactive state [11]. Obviously, the synaptic weights affect the stability of the SNNs [22], but the traditional methods cannot well address the problem of synaptic weight infinite modulation.…”