Liquid Computing Efficiency as a Function of Neural Cell's Electrical Parameters

G.M. Wojcik and W.A. Kaminski (Poland)


Hodgkin-Huxley neurons, visual system modelling, neural models parallelisation


For this study systems of Hodgkin Huxley neural networks consisting of about 16 thousands of cells are examined. We base on Liquid Computing theory. The separation abil ity of the system of Liquid State Machines depends on the choice of time constants characterising particular neu rons. The main objective of research presented in this arti cle is to investigate how this separation ability depends on the membrane resistance and its capacitance varying for a given time constant. Finding optimal ranges of parameters allows us to construct more effective liquid-based neural networks. Parameter search simulations and solving large number of nonlinear differential equations are always time and power consuming. That is why parallelisation of the problem is a demand. The results of parallel simulations are discussed. We propose a method of parallelisation for the network simulating a part of mammalian cortex and simple benchmark results for the local cluster are presented in some detail.

Important Links:

Go Back