Identifying Sub-networks in Neural Networks via Functionally Similar Representations