Abstract
In recent years, Spiking Neural Networks (SNNs) have demonstrated greatsuccesses in completing various Machine Learning tasks. We introduce a methodfor learning image features by \textit{locally connected layers} in SNNs usingspike-timing-dependent plasticity (STDP) rule. In our approach, sub-networkscompete via competitive inhibitory interactions to learn features fromdifferent locations of the input space. These \textit{Locally-Connected SNNs}(LC-SNNs) manifest key topological features of the spatial interaction ofbiological neurons. We explore biologically inspired n-gram classificationapproach allowing parallel processing over various patches of the the imagespace. We report the classification accuracy of simple two-layer LC-SNNs on twoimage datasets, which match the state-of-art performance and are the firstresults to date. LC-SNNs have the advantage of fast convergence to a datasetrepresentation, and they require fewer learnable parameters than other SNNapproaches with unsupervised learning. Robustness tests demonstrate thatLC-SNNs exhibit graceful degradation of performance despite the random deletionof large amounts of synapses and neurons.