Echo State Property of Neuronal Cell Cultures
- 2.9k Downloads
Physical reservoir computing (PRC) utilizes the nonlinear dynamics of physical systems, which is called a reservoir, as a computational resource. The prerequisite for physical dynamics to be a successful reservoir is to have the echo state property (ESP), asymptotic properties of transient trajectory to driving signals, with some memory held in the system. In this study, the prerequisites in dissociate cultures of cortical neuronal cells are estimated. With a state-of-the-art measuring system of high-dense CMOS array, our experiments demonstrated that each neuron exhibited reproducible spike trains in response to identical driving stimulus. Additionally, the memory function was estimated, which found that input information in the dynamics of neuronal activities in the culture up to at least 20 ms was retrieved. These results supported the notion that the cultures had ESP and could thereby serve as PRC.
KeywordsNeuronal cell culture Physical reservoir computing Echo state property Memory capacity
This paper is based on results obtained from a project (Exploration of Neuromorphic Dynamics towards Future Symbiotic Society) commissioned by NEDO, KAKENHI grant (17K20090), AMED (JP18dm0307009) and Asahi Glass Foundation. We thank Hitachi UTokyo Laboratory, Hitachi, Ltd. for fruitful discussions. K. N. was supported by JST PRESTO Grant Number JPMJPR15E7, Japan and KAKENHI No. JP18H05472, No. 16KT0019, and No. JP15K16076.
- 10.Hales, C.M., Rolston, J.D., Potter, S.M.: How to culture, record and stimulate neuronal networks on micro-electrode arrays (meas). JoVE (J. Vis. Exp.) (39), e2056 (2010). https://doi.org/10.3791/2056
- 11.Jaeger, H.: Identification of behaviors in an agent’s phase space. Citeseer (1995)Google Scholar
- 12.Jaeger, H.: Short term memory in echo state networks, vol. 5. GMD-Forschungszentrum Informationstechnik (2001)Google Scholar
- 13.Jaeger, H.: Tutorial on training recurrent neural networks, covering BPPT, RTRL, EKF and the “echo state network” approach, vol. 5. GMD-Forschungszentrum Informationstechnik Bonn (2002)Google Scholar
- 22.Nakajima, K., Hauser, H., Kang, R., Guglielmino, E., Caldwell, D.G., Pfeifer, R.: Computing with a muscular-hydrostat system. In: 2013 IEEE International Conference on Robotics and Automation, pp. 1504–1511. IEEE (2013). https://doi.org/10.1109/ICRA.2013.6630770