While I understand the OP's issue with the paper, I also feel that there is scope for the "We tried this and the improvement we got was minimalistic, so you sho

While I understand the OP's issue with the paper, I also feel that there is scope for the "We tried this and the improvement we got was minimalistic, so you should probably try a different approach" kind of paper. But OTOH, I agree that with the current "hype" around Deep learning, accompanied by the beginning of an "DL winter" in revolutionary papers means that academicians and companies which are set up in a "publish or perish" state of mind end up in a rush to publish even the smallest of modifications/enhancements. I understand that I'm arguing both sides of the table here, but at the end of the day I'd rather have these papers published than not, as long as they end up in public domain and can somehow be viewed more as experimental papers than purely theoretical ones.