UCNN unites ANN and SNN with magnitude-phase neural processing

A new paper from Reza Ahmadvand, Sarah Safura Sharif, and Yaser Mike Banad proposes the Unified Complex-valued Neural Network (UCNN), a computational model that bridges the gap between artificial neural networks (ANNs) and spiking neural networks (SNNs). The core innovation is the Unified Complex-valued Neuron (UCN), which operates with an asymmetric complex-valued state: the magnitude encodes continuous signal strength (like ANNs), while the phase governs intrinsic temporal dynamics and event-driven spike generation (like SNNs). This allows the model to learn both value encoding and timing within a single structure, addressing a long-standing limitation in neuromorphic computing.

The authors develop a foundational training framework that combines standard backpropagation for magnitude pathways with backpropagation through time (BPTT) for phase pathways. To reduce computational overhead, they introduce an event-driven adaptive phase learning (EAPL) rule that efficiently updates phase parameters only when spikes occur. Experimental validation on object tracking and Lorenz attractor tasks shows that UCNN achieves accurate and stable spatiotemporal learning while maintaining sparse, event-driven computation. This makes UCNN particularly promising for edge-AI and neuromorphic hardware, where energy efficiency and real-time processing are critical.