Randomly Spiking Dynamic Neural Fields Driven by a Shared Random Flow
Dynamic Neural Fields (DNF) is a well studied mean field model introduced by Amari. It is commonly used for high level bio-inspired cognitive architecture modeling or as a module for autonomous bio-inspired robotics. In a previous work we studied the feasibility of a purely cellular hardware implementation of this model in a digital substratum. We introduced the randomly spiking dynamic neural fields which successfully reproduced the DNF model’s behavior with local and decentralized computations implemented on FPGA. The lateral synaptic weights are computed with a random propagation of binary information generated with a cellular array of pseudo random number cellular automata. More than half of the area utilization was dedicated to the random numbers generation. In this paper we investigate two ways of reducing the surface of random number generators while keeping a cellular architecture.
KeywordsNormalize Root Mean Square Error Bernoulli Trial Inhibitory Layer Implementation Area Spike Diffusion
- 5.Mi, Y., Fung, C.C.A., Wong, K.Y.M., Wu, S.: Spike frequency adaptation implements anticipative tracking in continuous attractor neural networks. In: Advances in Neural Information Processing Systems, pp. 505–513 (2014)Google Scholar
- 7.Cerda, M., Girau, B.: Bio-inspired visual sequences classification. In: From Brains to Systems - Brain-Inspired Cognitive Systems (2010)Google Scholar
- 10.Vazquez, R.A., Girau, B., Quinton, J.: Visual attention using spiking neural maps. In: The 2011 International Joint Conference on Neural Networks (IJCNN), pp. 2164–2171, 31 July–5 August 2011Google Scholar
- 12.de Vangel, B.C., Torres-Huitzil, C., Girau, B.: Spiking dynamic neural fields architectures on FPGA. In: 2014 International Conference on ReConFigurable Computing and FPGAs (ReConFig), pp. 1–6, December 2014Google Scholar