Stochastic Quantum Spiking Neural Networks with Quantum Memory and Local Learning

A team of researchers led by Jiechen Chen, Bipin Rajendran, and Osvaldo Simeone has published a novel paper on arXiv introducing Stochastic Quantum Spiking Neural Networks (SQSNN). This work directly addresses key limitations in the nascent field of hybrid quantum-neuromorphic computing. Previous quantum spiking models relied on classical memory mechanisms implemented on single qubits, requiring repeated measurements to estimate firing probabilities and depending on conventional backpropagation for training—a process ill-suited for noisy quantum hardware. The newly proposed Stochastic Quantum Spiking (SQS) neuron model fundamentally changes this by using multi-qubit quantum circuits to create a spiking unit with built-in quantum memory.

The technical breakthrough is twofold: first, the SQS neuron enables event-driven, probabilistic spike generation in a single shot during inference, a significant efficiency gain. Second, and perhaps more importantly, networks of these neurons (SQSNNs) can be trained using a hardware-friendly local learning rule, completely bypassing the need for global classical backpropagation. This local training is a major step toward making such models feasible on actual quantum processors where global operations are challenging. The paper demonstrates through experiments on conventional and neuromorphic datasets that SQSNNs outperform both prior quantum spiking neural networks and their classical counterparts when the total number of trainable parameters is held constant, suggesting a more efficient use of quantum resources for temporal data processing.