| name | stochastic-quantum-spiking-neural-networks |
| title | Stochastic Quantum Spiking Neural Networks with Quantum Memory and Local Learning |
| description | Novel stochastic quantum spiking (SQS) neuron model with multi-qubit quantum circuits for internal quantum memory, enabling event-driven probabilistic spike generation and hardware-friendly local learning without backpropagation. |
| authors | ["Jiechen Chen","Bipin Rajendran","Osvaldo Simeone"] |
| arxiv_id | 2506.21324 |
| categories | ["quantum-computing","neuromorphic-computing","spiking-neural-networks","quantum-machine-learning"] |
| date | 2026-07-23 |
| version | 1.0 |
Stochastic Quantum Spiking Neural Networks with Quantum Memory and Local Learning
Overview
This methodology proposes a novel stochastic quantum spiking (SQS) neuron model that addresses key limitations of existing quantum spiking models. The SQS neuron uses multi-qubit quantum circuits to realize spiking units with internal quantum memory, enabling event-driven probabilistic spike generation in a single shot during inference. Networks of SQS neurons (SQSNN) can be trained via hardware-friendly local learning rules, eliminating the need for global classical backpropagation.
Core Methodology
SQS Neuron Architecture
- Multi-qubit quantum circuits: Realize internal quantum memory mechanisms beyond single-qubit classical memory
- Single-shot inference: Event-driven probabilistic spike generation without repeated measurements
- Quantum memory integration: Internal state preservation through quantum coherence and entanglement
- Hardware-friendly design: Compatible with neuromorphic integrated sensing and communications (N-ISAC)
SQS Neural Networks (SQSNN)
- Local learning rules: Hardware-friendly training without global backpropagation
- Event-driven computation: Energy consumption only upon input events
- Tensor-product composition: Exponential state space growth with qubit count
- Superposition and entanglement: Quantum states across basis states and subsystems
Training Methodology
- Local learning rule: Eliminates need for conventional backpropagation
- Parameter efficiency: Improved performance when fixing total trainable parameters
- Hardware compatibility: Designed for implementation on quantum neuromorphic platforms
Applications
Neuromorphic Integrated Sensing and Communications (N-ISAC)
- Event-driven applications requiring real-time processing
- Low-power quantum neuromorphic systems
- Hybrid quantum-classical edge computing
Time Series Processing
- Efficient temporal data processing through sparse, event-driven computation
- Quantum-enhanced pattern recognition in time series
- Real-time decision making with quantum speedup
Quantum Machine Learning
- Parameter-efficient quantum neural networks
- Hardware-native quantum ML architectures
