It amazes me how many "high profile" researchers have blatantly fraudulent work. As an example, take this rather recent ICLR submission that is egregiously pla

It amazes me how many "high profile" researchers have blatantly fraudulent work. As an example, take this rather recent ICLR submission that is egregiously plagiarized[0] (I mean attempts to hide the plagiarism are minimal to nonexistent). Clicking on the authors you can view their profiles and you'll find that most authors have several thousand citations and one north of 25k[1]!!!! They also have an i-10 of 400, which is highly abnormal for that citation level while the last author has an i-10 of 126 with only 7.3k citations. The problem I see here is that there's lots of suspicious activity and clear evidence of fraud, yet little to no action is taken. If anything, it seems we mostly ignore it. The ICLR submission is a bit odd too, since decisions and comments are public, most of the time this happens behind closed doors and we don't observe the evidence. As far as I'm aware, no one is making network graphs to track fraudulent authors. I think we want to sweep it under the rug because at the end of the day our system relies heavily upon trust. Unless we strongly incentivize reproduction we require high trust and we're afraid to acknowledge fraud in the community because we believe it will make more distrust science. But I think in reality, the reluctance to pursue fraud and the reluctance to have high clarity creates more distrust. The truth is that this system poorly scales even before we incorporate the reverse (bureaucratic) incentives. I think we lost sight of the goals of research and publishing. Publishing exists to facilitate communication among our peers. We publish in venues for two reasons: to improve/error check works and to facilitate social networks. Publish or perish is detrimental as well as the push for highly subjective measures like novelty. I strongly advocate for returning to the old paradigm, where the conditions for publication are: are there any major errors, is there plagiarism, and does the work sufficiently provide evidence for the experimental hypothesis (note: I intentionally do not say "prove".). This system accepts the reality that we cannot sufficiently judge significance. Breakthroughs by definition come from research areas that are under studied. Frequently from directions that had previously been rejected. Unfortunately, this is becoming more common due to the speed in which researchers must publish and exacerbated by the rise in thresholds to get a work published. Many ideas are abandoned not due to evidence that they will not yield progress, but that the belief to achieve "sufficient" evidence would require too much time. While tenure should "resolve" this, it ignores the fact that a tenured professor still has students that must publish as well as they've been working in a very different way for years. And old habits are hard to break. I think we need to take a hard long look at our system and I suspect we could do so much more if we rebuilt the structure. [0] https://openreview.net/forum?id=cIKQp84vqN https://openreview.net/forum?id=cIKQp84vqN [1] https://scholar.google.com.hk/citations?user=D-TS1fAAAAAJ https://scholar.google.com.hk/citations?user=D-TS1fAAAAAJ