Image Classification using Convolution Liquid State Machines

A Convolutional Liquid State Machine (CLSM) combines Liquid State Machines (LSMs) and Convolutional Neural Networks (CNNs) for enhanced feature extraction and classification. CLSMs, based on spiking neural networks, handle temporal and high-dimensional data efficiently and are therefore suitable for real-time image classification. With the addition of convolutional layers, they improve spatial feature representation and reduce computational cost. This paper investigates CLSMs' versatility, tuning reservoir size, synaptic connectivity, and feature extraction. CLSMs are compared with CNNs and Transformers in terms of accuracy, recall, precision, F1-score, and efficiency. CLSMs offer energy-efficient, high-performance solutions to autonomous navigation, medical imaging, and edge AI, extending the boundaries of neuromorphic computing.