>>12103682>>12103682>It’s more accurate to say many and most are scrambling to get more and more empirical results because it’s what the shareholders demandI WILL say as someone familiar with the literature (including JMLR, love the journal setup, fuck elsevier), that half the papers are "we tried out a fuck-ton of architectures, this random subset seemed to work really well for this problem, lets figure out a mathematical reason why" and they subtely switch the story around so it seems like they logic'd their way to a best-practices for [insert data type]
We do it too in my area (cell biology), but its quite obvious sometimes
