Beyond dimensionality reduction and reducing computation cost, I am trying to imagine the use of one dimensional convolutions as feature detectors. Not really sure if that's something that even makes sense...THanks for the piece! As always enlightening! Cheers