Hacker News Reader

Double blind is a farce in practice in most CS specialties (maybe not bloated fields like AI/HCI/graphics?). Once you get to the top tier, everyone knows everyone, including the industrial labs, including what they're working on. Most good work is too differentiated to anonymize to well-versed readers: less Neural Networks for X, more Sub-Theory of a Sub-Theory or Version 2 of This Weird Thing.

This may not be obvious to the normal grad student, but after a decade in a community of 1K-10K people, most of them working in groups, it becomes clear. The end result is that the last several top-tier conferences I reviewed for, I basically guessed the authors of all the papers I reviewed, and knew at least one of them personally on each paper. All of that, despite being a "young" member of the community.

> maybe not bloated fields like AI/HCI/graphics

Having reviewed papers in the 'bloated' graphics field at SIGGRAPH, yes, you can usually guess who wrote a particular paper.

> This may not be obvious to the normal grad student, but after a decade in a community of 1K-10K people

The communities are smaller than that. Within sub-fields in graphics, you usually have perhaps five or six research groups to guess from.

> The CS field gets a lot of bashing for gravitating a lot around conferences rather than journals. Because, you know, journals are supposed to be the serious venue for the grown-ups.

I just remember how my university, who had some standard and fairly comprehensive subscriptions to electronic IEEE and ACM journals, didn't get conference proceedings for some damn reason. From the critically-thinking student's point of view, conferences and journals look like two different ways of getting students to part with money with no recourse if the product happens to suck, as academic material so often does.

"From the critically-thinking student's point of view, conferences and journals look like two different ways of getting students to part with money with no recourse if the product happens to suck, as academic material so often does."

I can't speak to journals, but as someone who's been on both sides of IEEE conferences, that it absolutely true. I've seen conferences where almost none of the researchers showed up to talk about their work---the students had to get published to graduate, so they submitted, got accepted, paid their registration fees (literally, the most important step), and, badaboom, they were officially published.

My personal favorite was ICDCS '98[1], where early student conference registration included a copy of the proceedings, until you got to the conference, where you were informed that, of course students couldn't get free copies of the proceedings. What were you thinking?

[1] The conference had a serious hiccup, committee-wise, and had to change venues at the last moment, so....

Would you mind explaining the double-blind review in CS conferences? I'm a doctor by trade so am well acquainted with the way peer-review works but I've never heard of conferences being reviewed, either before-the-fact as an acceptance criterion or after-the-fact as a kind of rating.

CS conference papers go through the same process as a journal would in your field, but typically with only 1-2 revision cycles. The acceptance rates are much lower than journal because there are fewer retries, so people generally submit ready-to-publish work to conferences. Many researchers don't publish in journals at all, or just do it as a token activity.

In the context of peer review (at least in CS), "single-blind" means that the authors don't know who the reviewers are, and "double-blind" means that additionally, the reviewers don't know who the authors are.

Conferences with double-blind review require papers to be anonymized, avoiding any mentions of author names or affiliations in the PDF submitted for review, as well as other obvious elements that would reveal the identity of the authors (e.g. "As we showed in a previous paper [4], ...")

"..."single-blind" means that the authors don't know who the reviewers are..."

Technically, it means that the authors would have to do some minor text analysis to find out who the reviewers are; the only difficulty is that the comments can be a bit thin, but you've got plenty of known samples to measure against.