rasbt
commented
3 years ago I agree. It's very hard to find these papers though. The problem is usually that papers that involve deep learning are centered around deep learning, and the authors try to show that it outperforms other methods.
One recent paper that comes to mind was related to tabular datasets, i.e., here one take-away was that DL works well on tabular data, but non-DL methods (here: gradient boosting) can perform similarly well: https://arxiv.org/abs/2106.11959. The one problem here is though that gradient boosting is not necessarily simpler than DL
Saw a study that supports one of our claims re ML baselines and added it