Neuromorphic and Bio-Inspired Learning Strategies for Sustainable and Low-Power AI

In the contemporary digital era, the proliferation of intelligent systems is increasing exponentially, challenging the capabilities of conventional computing paradigms. As an increasing number of applications migrate to the edge, the necessity for sustainable and low-power solutions has become increasingly urgent. This demand has contributed to the advancement of neuromorphic computing, a field inspired by the principles of biological neural processing. Neuromorphic computing offers an energy-efficient, online learning and adaptation alternative to traditional systems, especially under strict resource constraints. Concurrently, biologically plausible and backpropagation-free approaches are emerging as complementary strategies that draw inspiration from the brain learning mechanisms, avoiding implausible mechanisms of traditional deep models. Neuromorphic and biological plausible approaches are shifting Artificial Intelligence (AI) towards more natural, adaptive, and sustainable systems. This thesis explores neuromorphic and biologically plausible learning strategies, with a focus on their application in the design of low-power, sustainable intelligent systems. The investigation encompasses Spiking Neural Networks (SNNs), event-driven processing, and local plasticity-based rules. By comparing these paradigms across different application domains, the work highlights trade-offs in accuracy, latency, power consumption, and scalability. Experimental results demonstrate how carefully designed algorithms can achieve significant gains in efficiency without compromising performance, paving the way toward the next generation of intelligent, autonomous systems.