Integrated feature and parameter optimization for an evolving spiking neural network: exploring heterogeneous...

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Bio-inspired Computing

Background:

• Spiking neural networks (SNNs) offer efficient, biologically plausible computation.
• Optimizing SNNs, including their features and parameters, remains a significant challenge.
• Quantum-inspired algorithms can enhance complex optimization tasks.

Purpose of the Study:

• To introduce a novel quantum-inspired spiking neural network (QiSNN) for integrated optimization.
• To develop a hybrid optimization approach using distinct representations for features and network parameters.
• To evaluate the performance of the QiSNN on benchmark and real-world datasets.

Main Methods:

• Developed a QiSNN integrating feature and parameter optimization using a quantum-inspired evolutionary algorithm.
• Employed a dual-representation strategy: binary for feature selection and continuous for parameter tuning.
• Validated the framework on synthetic datasets and a real-world ecological dataset for invasive species prediction.

Main Results:

• The QiSNN demonstrated significantly faster convergence towards optimal solutions compared to traditional methods.
• Achieved superior accuracy in classification tasks.
• Identified a more informative subset of features, enhancing model interpretability.

Conclusions:

• The proposed QiSNN offers an effective and efficient approach for optimizing spiking neural networks.
• The dual-representation optimization strategy enhances both performance and feature selection.
• This framework shows promise for applications in complex prediction tasks, including ecological modeling.