luciaquirke comments on Deep sparse autoencoders yield interpretable features too