Synaptic Plasticity Neural Modeling: Dissimilar Memristor Circuits.
Synaptic plasticity modeling may have another boost forward with a recent paper proposal from two Eastern researchers. In “Bottleneck of using single memristor as a synapse and its solution”, researchers Farnood Merrikh-Bayat and Saeed Bagheri Shouraki have stated they were able to more closely model a Hebbian synaptic plasticity model by demonstrating the phenomena under a dissimilar memristors-in-serial circuit. Under Hebbs Rule, the generalization of synaptic strength relationships (“Neurons that fire together, wire together”) in determining weight alterations between model neurons (“synaptic strengthening”), has not been an easy relationship to model physically. But by utilizing two dissimilar memristors in series, the ability to model a basic Hebbian learning relationship is improved:
Now, it is well known that one of the main applications of memristor is for the hardware implementation of synapses because of their capability in dense fabrication and acting as a perfect analog memory. However, synapses in biological systems have this property that by progressing in the learning process, variation rate of the synapses weights should decrease which is not the case in the currently suggested memristor-based structures of neuromorphic systems. In this paper, we show that using two dissimilar memristors connected in series as a synapse perform better than the single memristor. [arxiv.org abstract] [paper v1 (pdf)]
Its a tribute to the early work of both Donald Hebb and Norbert Wiener in neural modeling that the field may be able to advance beyond slime molds before mid-century. More complex models than the range of spike timing dependent plasticity have implications across a broad range of research applications.