Dr Rakesh Sengupta, Assistant Professor, Psychology, SIAS’ latest paper titled Bio-Inspired Variance Control: Decoupling Signaling Mode from Power Constraints in Neuromorphic Systems has been published and presented at the 2026 IEEE International Conference on Interdisciplinary Approaches in Technology and Management for Social Innovation (IATMSI) in Gwalior.
The paper tackles the problem statement on how as smart devices and the Internet of Things (IoT) grow, they consume massive amounts of power, and Current AI systems struggle to balance high performance with energy efficiency. The research for this study draws inspiration from how the human cerebral cortex operates, this paper introduces a new mathematical model called “Homeostatic Variance Control. By mimicking the brain’s natural mechanisms, we demonstrated a way for systems to seamlessly transition into highly sparse, energy-efficient states without altering their baseline power budget. This provides a rigorous new framework for designing sustainable, low-power AI for the future of cyber-physical systems.
Abstract
Advanced intelligent systems and IoT networks face a critical trade-off between signaling flexibility and power efficiency. Traditional rate-coding schemes often require shifting the mean activity to transmit information, incurring high energy costs that threaten system stability. Drawing inspiration from the Log-Normal firing regimes of the cerebral cortex, we propose a novel coding mechanism: Homeostatic Variance Control. We mathematically demonstrate that by modulating the variance of input noise (a proxy for neuromorphic gain), a system can dramatically shift its modal (most frequent) operating point toward sparsity while pinning its median resource consumption to a fixed homeostatic setpoint. We validate this mechanism using a Linear-Nonlinear-Poisson (LNP) simulation, showing that it enables a seamless transition between dense and sparse coding regimes without altering the baseline power budget. This bioinspired architecture offers a rigorous framework for designing resilient, energy-efficient Spiking Neural Networks (SNNs) for resource-constrained cyber-physical applications.
