Demian Till comments on Do sparse autoencoders find “true features”?

[Demian Till]

Thanks, I mentioned this as a potential way forward for tackling quadratic complexity in my edit at the end of the post.