Learnable Spiking Neural P System with Interval Excitation

ICLR 2026 Conference Submission12864 Authors

Keywords: Spiking Neural P systems, parallel distributed models, interval excitation, surrogate gradients, spiking computations, hardware-adapted learning

Abstract: Spiking Neural P (SN P) systems are parallel distributed models developed by mimicking bio-nervous systems. Past decades have emerged a lot of efforts on theoretical characterizations and modeling plasticity of SN P systems; however, it still remains challenging that interacts with real-world environments due to the limited expressive capacity and the non-differentiable nature of their excitation mechanism. This paper proposes a Learnable Spiking Neural P System with Interval Excitation (LSNP_IE) for real-valued processing. The proposed LSNP_IE employs an interval excitation mechanism and a potential adjustment module, which improve modeling plasticity and enable excitation stability, respectively. The whole system can be adjusted by surrogate gradients beyond hardware. Experimental results conducted on real-world datasets show that LSNP_IE achieves competitive performance compared to traditional non-spiking and spiking models. Our investigations not only reveal the potential of integrating spiking computations with parallel distributed frameworks, but also support the development of hardware-adapted learning.

Primary Area: applications to neuroscience & cognitive science

Submission Number: 12864