Enhancing Neuromorphic Robustness via Recurrence Resonance: The Role of Shared Weak Attractors in Quantum Logic Networks

Recurrence resonance, a phenomenon that enhances system computational capability by exploiting noise to amplify hidden attractors, holds significant potential for applications such as edge computing and neuromorphic computing. Although previous studies have extensively explored its characteristics, the underlying mechanism regarding its generation remains unclear. Here, we employed a Stochastic Recurrent Neural Network to simulate neural networks under various coupling conditions. By introducing appropriate inhibitory connections and examining the state transition matrices, we analyzed the characteristics and correlations of attractor landscapes in both global and local systems to elucidate the generative mechanism behind the "Edge of Chaos" dynamics observed under the quantum logic connectivity structure during recurrence resonance. The results show that the strategic introduction of inhibitory connections enriches the system's attractor landscape without compromising the intensity of recurrence resonance. Furthermore, we find that when neurons are coupled via quantum logic and noise intensity meets specific conditions, the strong attractors of the global system decompose into those of distinct local subsystems, accompanied by the sharing of structurally similar weak attractors. These findings suggest that under quantum logic connectivity, the interaction between the strong attractors of different subsystems is mediated by a background of shared weak attractors, thereby enhancing both the system's robustness against noise and the diversity of its state evolution.