[AussieWog93]

Somewhat tangential, but when I was doing my PhD there was a bizarre amount of research thrown into testing ML algorithm after ML algorithm at bad EEG data in the hope that we'd be able to magically find a signal from noisy garbage input data.

"Progress" consisted of someone finding a new algorithm that just so happened to get good performance on one particular dataset (but not others).

Everyone knew it was bullshit but did it anyway, because it was easy to convince people to give you grants if you have a sexy, sellable hypothesis and a willingness to handwave away the two decades of prior non-progress.