Building a mathematical model for generic neural microcircuits
Reservoir computing networks have recently received attention as a potential mathematical model for generic neural microcircuits. They also show promise as a computational technique for technical applications such as time series prediction, however they suffer significant variation in performance when applied to different problems and require the development of problem-specific optimisation procedures. We have investigated methods for addressing this issue in collaboration with researchers in Japan and Taiwan, and our findings have recently been published in the Human Frontier Science Program Journal. Our results have both potential for improving engineering applications, as well as implications for understanding how biological neural networks are shaped to be optimally adapted to the requirements of their environment.
Our results suggest methods for improving the predictive and memory capacity of Reservoir Computing networks such as this.
Joschka Boedecker, Oliver Obst, N. Michael Mayer, and Minoru Asada. "Initialization and self-organized optimization of recurrent neural network connectivity." HFSP Journal. Volume 3, Issue 5, pp. 340-349.
The paper is available here
Warning: may contain traces of Maths
Read more about CSIRO's research in Adaptive Systems.
CSIRO contact:
