Lucas Beyer at Meta argues that machine learning paper authors must examine similar prior work in detail to specify how their contributions differ from existing studies
Leo Boytsov replied that exhaustive reading is often unrealistic under publishing pressures.
@giffmana @stanislavfort I remember being surprised in college finding many ML papers citing Vapnik's 1968 "Soviet Mathematics Doklady" paper for uniform convergence results .... did they really read it? It was in Russian and not in any western databases
Lucas Beyer (bl16)@giffmana@stanislavfort wtf? "some people did something similar in this space" is literally the papers you SHOULD read in detail and explain exactly how what you do differs or complements! It's your job as author, not the reader's job.
@giffmana @stanislavfort There are a lot of publishing authors are expected to do in theory, but in practice it is often not feasible. Who reads papers in full nowadays? You only need to read enough to understand the difference. PS: I think WTF isn't conducive to a healthy discussion.
Lucas Beyer (bl16)@giffmana@stanislavfort wtf? "some people did something similar in this space" is literally the papers you SHOULD read in detail and explain exactly how what you do differs or complements! It's your job as author, not the reader's job.