Abstract
The dynamics of signal transmission in neuronal networks remain incompletely understood. In this study, we propose a novel Rulkov neuronal network model that incorporates Q-learning, a reinforcement learning method, to establish efficient signal transmission pathways. Using a simulated neuronal network, we focused on a key parameter that modulates both the intrinsic dynamics of individual neurons and the input signals received from active neighbors. We investigated how variations in this parameter affect signal transmission efficiency by analyzing changes in attenuation rate, as well as the maximum and minimum firing intervals of the start and goal neurons. Our simulations revealed that signal transmission efficiency between distant neurons was significantly impaired in the parameter region, where a chaotic attractor and an attractor of the eight-periodic points are observed to co-exist. A key finding was that low-frequency oscillatory bursts, while failing long-distance transmission, were capable of amplifying signals in neighboring neurons. Furthermore, we observed variation in signal transmission even when individual neuron dynamics remained similar. This variability, despite similar presynaptic activity, is a biologically significant phenomenon, and it is argued that it may contribute to the flexibility and robustness of information processing. These findings are discussed in the context of their biological implications.