Optimize the Spiking Neural Network

Objective

Reduce the memory footprint of our simulated 1-million neuron SNN by 40% without losing spike accuracy. Spiking Neural Networks (SNNs) are promising for low-power AI, but scaling them to millions of neurons requires massive memory bandwidth. Your task is to implement a sparse representation technique that compresses the synaptic weight matrices while maintaining at least 95% of the original classification accuracy on the provided benchmark dataset.

Requirements

• Python 3.9+
• PyTorch or NumPy
• Memory constraint: < 2GB RAM