Wannabe computer science superstars must all run the same rather scary and capricious gauntlet, one that sounds deceptively dull: the computer science conference paper review process. To have a research paper accepted for presentation at a CS conference is a coveted rite of passage among academics and professionals, bestowing on its author a status symbol that can open the door to tenure or competitive job offers.
Last month, the University of California, Berkeley’s much-respected Edward Lee, a professor emeritus of electrical engineering and computer sciences who for several decades has served on program committees that judge research papers, caused an uproar in the CS community after he publicly shared a scathing review of the system, which he’d sent earlier to fellow judges. Program committee members who decide which papers are accepted are volunteers, members of the academic community who agree to spend hours of their time (theoretically) reading submissions, writing opinions and voting on whether papers are worthy of the hallowed halls of whatever conference is in session.
But ever since conferences adopted a new review process that shields the names of judges, as well as papers’ authors — many made the change in the early 2000s — critics say a new problem has arisen: Rejection notes are often so random, or just factually incorrect, that applicants suspect nobody actually read their paper. The issue came to a head for Lee earlier this year. Here is his story of why he’s chastising the community he’s been a loyal member of for so long, and what he thinks can be done to address the problem.
Lee’s story, as told to Protocol, has been edited for clarity and brevity.
It started when I was serving on a program committee for one of my favorite conferences. It was that program committee experience that pushed me over the edge, and I wrote a letter to the entire program committee and resigned.
I had found myself fighting with a lot of the program committee members over determinations about papers, and a number of papers that I thought were highly worthy got rejected. There were two papers that were principally authored by students that I had worked with closely, and of course I couldn't participate in the deliberations because of conflict of interest rules. But those papers got rejected with what I considered to be unsound reviews. And I have enough experience to know that these two papers were excellent papers.
My resignation and protest letter got quite a few people upset. I got quite a bit of feedback. It has become very clear to me that there are a lot of people who are very frustrated with the current situation. This piece seems to have really resonated with a lot of people because everyone in the community is facing 10% acceptance rates for their papers. I have seen some extremely talented people leave the field because of brutal reviews. And that's just unacceptable.
Employers need to know the reality is that getting conference papers accepted is extremely random. Looking at published conference papers in computer science as a measure of the quality of the candidate is flawed. You're looking at luck. If you want to hire lucky people, OK. That’s usually not what they're looking for.
And then the Sigbed blog editors somehow got wind of my open letter to the program committee and asked me if I would submit it as a blog, which is how the Sigbed blog post came about.
I've been doing these kinds of reviews for my entire career. So that's 40 years. I get invited to be on a lot of these program committees, but I simply don't have the bandwidth for them. So I typically serve on two or three a year, trying to pick the type of conferences that I can contribute the most to.
The problem has been there all along, but it was much less visible to me because the reviews didn't use to be double-blind. The students that I worked with the most closely were almost always from Berkeley, and Berkeley papers weren't rejected as often as papers from other places.
So in some ways, the institution of double-blind review processes has been a very good thing, because there were prejudices creeping into the review process unknowingly. The papers from the best institutions were more likely to be accepted, papers written by males rather than females were more likely to be accepted. Papers with Chinese names were more likely to be rejected. The double-blind review process put an end to that problem.
But that also exposed to me the high rejection rates and the frustration that accompanies them, because the reviews are frankly capricious and often unsound.
Part of the problem is that the program committees are being asked to do more than is actually possible. In the past, they could rely on a kind of a crutch. It’s an MIT paper, it's probably pretty good. Let's just accept it. But they can't do that anymore.
There’s also the anonymity. There are good reasons for keeping the reviewers anonymous — you don't want junior people who are reviewing to be vulnerable to retribution from senior people who get their papers rejected. But people can be much more mean when they are anonymous. And moreover, when you combine that with the fact that reviews themselves never get published, their critique of that paper is protected.
One thing that we could do that would improve things quite a bit is keep the double-blind process, but the original submission and the reviews get published, right along with the paper. That way the conference gets associated with the reviews, and if the conference has a lot of capricious reviews, that's going to degrade the reputation of the conference. Right now there’s basically a lot of power with no accountability, which is almost never a good thing.
The first open letter that I sent got circulated to all the new program committee members in another related conference just shortly thereafter. I've seen quite a bit of discussion about being a lot more careful about using novelty as a criteria for rejection, for example, which is one of the things I argue against in this blog.
I'm hoping that there will be some impact. I've been collecting notes from all the feedback I've been getting, and I might have enough to put together a more upbeat follow-up blog that discusses some real concrete actions that can be taken.
🗣 How I Decided… 🗣
Wednesday, June 29
Wednesday, July 6
Wednesday, July 13
Wednesday, July 20
Wednesday, July 27
Wednesday, Aug. 3
Wednesday, Aug. 10
Wednesday, Aug. 17
Wednesday, Aug. 24
Wednesday, Aug. 31
Wednesday, Sept. 7
Wednesday, Sept. 14
Wednesday, Sept. 21
Wednesday, Sept. 28
Wednesday, Oct. 12
Wednesday, Oct. 19
Wednesday, Oct. 26
Wednesday, Nov. 2