Neuron Pruning Experiments

This repository contains experiments on neuron pruning. The goal is to compare different approaches for pruning neurons and weights in neural networks and evaluate their performance.

Overview

The repository implements and compares three approaches:

1. Neuron-level Popup Model: Prunes entire neurons using a score-based selection method inspired by edge-popup.
2. Weight-level Popup Model (Edge-Popup): Prunes individual weights using a score-based selection method.
3. Standard MLP: A baseline fully connected network trained without pruning.

All models typically share the same architecture (2 layers with the same number of neurons) and are evaluated on training and test accuracy.