Read on Twitter
The T5 paper has been published in JMLR! 
http://jmlr.org/papers/v21/20-074.html
Since I have already talked more than enough about T5, here instead is a thread about the (awesome) process of publishing in JMLR:
(1/10)
http://jmlr.org/papers/v21/20-074.html
Since I have already talked more than enough about T5, here instead is a thread about the (awesome) process of publishing in JMLR:
(1/10)
When we were working on the T5 paper, it was clear that it was going to be much too long to be a conference paper. It was important to some of the authors that the paper not be arxiv-only, so we decided to shoot for a journal publication.
(2/10)
(2/10)
JMLR was an obvious choice because it
- is 100% free (as in libre *and* gratis)
- is well-respected
- accepts comically long papers
(if you aren't familiar with the very cool genesis of JMLR, see http://www.jmlr.org/statement.html )
(3/10)
- is 100% free (as in libre *and* gratis)
- is well-respected
- accepts comically long papers
(if you aren't familiar with the very cool genesis of JMLR, see http://www.jmlr.org/statement.html )
(3/10)
The first thing that surprised me about submitting to JMLR is that you suggest your own reviewers! Conflicts of interest aside, this is makes a lot of sense - you, as an author, probably know who the best-qualified people to read your paper are.
(4/10)
(4/10)
In contrast, submissions to current ML conferences often end up with a somewhat-random group of questionably-qualified and disgruntled reviewers who are simultaneously reviewing 8+ papers under time pressure.
(5/10)
(5/10)
I think many people shy away from submitting to journals because they think that the review process takes an extremely long time. Surprisingly, our reviews came back *faster* than the typical ML conference review period!
(6/10)
(6/10)
Even better, the reviews were all genuinely helpful and constructive and I truly believe that the camera-ready version is meaningfully improved thanks to the reviewer's comments. Who would have thought such a thing was possible?
(7/10)
(7/10)
There are many other benefits that I didn't mention above:
- No submission or conference registration fees
- No travel/carbon emissions for going to the conference
- Paper is published immediately when it is given the final go-ahead
- No "anonymity period"
etc. etc.
(8/10)
- No submission or conference registration fees
- No travel/carbon emissions for going to the conference
- Paper is published immediately when it is given the final go-ahead
- No "anonymity period"
etc. etc.
(8/10)
Finally, I'll point out that journals avoid other pathologies of conferences, like:
- all papers being exactly 8 pages (even though many would benefit from more experiments, explanation, and exposition)
- the huge rush to finish papers around arbitrary deadlines
etc.
(9/10)
- all papers being exactly 8 pages (even though many would benefit from more experiments, explanation, and exposition)
- the huge rush to finish papers around arbitrary deadlines
etc.
(9/10)
I definitely will be submitting more papers to JMLR in the future and I encourage you to give it some consideration for your next mega-project.
(10/10)
(10/10)
