Fuzzy leaky integrate-and-fire clustering algorithm

This paper presents a novel unsupervised clustering algorithm that combines incremental fuzzy c-means clustering with a discrete Leaky Integrate-and-Fire (LIF) neuron model. The method derives and updates the synaptic weights of LIF neurons by optimising a clustering objective function, facilitating incremental and fully unsupervised learning. Cluster formation occurs through spike-count-based neuron responses, with each neuron representing a prototype of a cluster. The effectiveness of the proposed approach is demonstrated using both synthetically generated datasets and benchmark datasets from the UCI Machine Learning Repository. Clustering performance is assessed using confusion matrices and normalised mutual information (NMI), showing that it is competitive with and in some instances superior to, conventional clustering techniques. The proposed algorithm exhibits several advantageous properties, including scalability to large datasets, computational efficiency, structural similarity to classical clustering methods, adaptability to new datasets and suitability for neuromorphic implementation, which provides benefits in energy efficiency, processing speed and fault tolerance.