Self-management for Neural Dynamics in Brain-like Information Processing

摘要

Neural dynamics coupled by adaptive synaptic information transmission provide a very powerful tool for biologically inspired visual processing systems[4]. Currently, progress is limited by the computing time needed to evaluate the underlying equations and by the high number of parameters necessary to tune to achieve the desired system performance.rnIn this contribution we apply Autonomic Computing techniques to overcome these limitations. We approach the computing time problem with an error model of the differential equations allowing for self-optimization of the evaluation step size and the parameter problem with a self-configuration heuristics to keep neural activation in working range.rnWe show the equivalence of system behavior compared to the case without self-management, the performance gain achieved by the self-optimization and the stability achieved by the self-configuration.