Do Neural Networks Trained with Topological Features Learn Different Internal Representations?