Representation Alignment In Neural Networks