Topological Insights in Sparse Neural Networks