During a research conversation this week I came to the realization that we really should model the paper writing as a partially observed Markov decision problem (POMDP). The writer has an action space of writing SIMPLE or FANCY statements. A SIMPLE statement follows very clearly from the previous statement and so should be easy to follow. A FANCY statement is a bit more challenging but might sound more impressive the reviewer. The reviewer has several states: BORED, IMPRESSED, and CONFUSED. Based on the input being SIMPLE or FANCY, the reviewer changes state according to a Markov chain with this state space.
The writer suffers a cost (or gains a reward, depending on how we want to model it) based on the current action and reviewer state. For example, issuing a FANCY statement when the reviewer is CONFUSED might suffer a higher cost than a FANCY statement when IMPRESSED. The goal is to minimize the discounted cost. This makes sense since reviewers become more fatigued over time, so extracting reward earlier in the paper is better.
Now, conventional wisdom might suggest that clarity is best, so a pure-SIMPLE policy should be a dominant strategy for this problem. However, depending on the cost and transition structure, I am not sure this is true.
And now, back to paper writing for me…
Why oh why can’t we have nice web-based software for academic things?
For conferences I’ve used PaperCept, EDAS (of course), Microsoft’s CMT, and EasyChair. I haven’t used HotCRP, but knowing Eddie it’s probably significantly better than the others.
I can’t think of a single time I’ve used PaperCept and had it work the way I expect. My first encounter was for Allerton, where it apparently would not allow quotation marks in the title of papers (an undocumented restriction!). But then again, why has nobody heard of sanitizing inputs? The IEEE Transactions on Automatic Control also uses PaperCept, and the paper review has a character restriction on it (something like 5000 or so). Given that a thorough review could easily pass twice that length, I’m shocked at this arbitrary restriction.
On the topic of journal software, the Information Theory Society semi-recently transitioned from Pareja to Manuscript Central. I have heard that Pareja, a home-grown solution, was lovable in its own way, but was also a bit of a terror to use as an Associate Editor. Manuscript Central’s editorial interface is like looking at the dashboard of a modern aircraft, however — perhaps efficient to the expert, but the interaction designers I know would blanche (or worse) to see it.
This semi-rant is due to an email I got about IEEE Collabratec (yeah, brah!):
IEEE is excited to announce the pilot rollout of a new suite of online tools where technology professionals can network, collaborate, and create – all in one central hub. We would like to invite you to be a pilot user for this new tool titled IEEE Collabratec™ (Formerly known as PPCT – Professional Productivity and Collaboration Tool). Please use the tool and tell us what you think, before we officially launch to authors, researchers, IEEE members and technology professionals like yourself around the globe.
What exactly is IEEE Collabratec?
IEEE Collabratec will offer technology professionals robust networking, collaborating, and authoring tools, while IEEE members will also receive access to exclusive features. IEEE Collabratec participants will be able to:
* Connect with technology professionals by location, technical interests, or career pursuits;
* Access research and collaborative authoring tools; and
* Establish a professional identity to showcase key accomplishments.
Parsing the miasma of buzzwords, my intuition is that this is supposed to be some sort of combination of LinkedIn, ResearchGate, and… Google Drive? Why does the IEEE think it has the expertise to pull off integration at this scale? Don’t get me wrong, there are tons of smart people in the IEEE, but this probably should be done by professionals, and not non-profit professional societies. How much money is this going to cost? The whole thing reminds me of Illinois politics — a lucrative contract given to a wealthy campaign contributor after the election, with enough marketing veneer to avoid raising a stink. Except this is the IEEE, not Richard [JM] Daley (or Rahm Emmanuel for that matter).
As far as I can tell, the software that we have to interact with regularly as academics has never been subjected to scrutiny by any user-interface designer. From online graduate school/faculty application forms (don’t get me started on the letter of rec interface), conference review systems, journal editing systems, and on, we are given a terrible dilemma: pay exorbitant amounts of money to some third party, or use “home grown” solutions developed by our colleagues. For the former, there is precious little competition and they have no financial incentive to improve the interface. For the latter, we are at the whims of the home code-gardener. Do they care about user experience? Is that their expertise? Do they have time to both make it functional and be a pleasure to use? Sadly, the answer is usually no, with perhaps a few exceptions.
I shake my fist at the screen.
The most popular topic of conversation among information theory afficionados is probably the long review times for the IEEE Transactions on Information Theory. Everyone has a story of a very delayed review — either for their own paper or for a friend of theirs. The Information Theory Society Board of Governors and Editor-in-Chief have presented charts of “sub-to-pub” times and other statistics and are working hard on ways to improve the speed of reviews without impairing their quality. These are all laudable. But it occurs to me that there is room for social engineering on the input side of things as well. That is, if we treat the process as a black box, with inputs (papers) and outputs (decisions), what would a machine-learning approach to predicting decision time do?
Perhaps the most important (and overlooked in some cases) aspects of learning a predictor from real data is figuring out what features to measure about each of the inputs. Off the top of my head, things which may be predictive include:
- number of citations
- number of equations
- number of theorems/lemmas/etc.
- number of previous IT papers by the authors
- h-index of authors
- membership status of the authors (student members to Fellows)
- associate editor handling the paper — although for obvious reasons we may not want to include this
I am sure I am missing a bunch of relevant measurable quantities here, but you get the picture.
I would bet that paper length is a strong predictor of review time, not because it takes a longer time to read a longer paper, but because the activation energy of actually picking up the paper to review it is a nonlinear function of the length.
Doing a regression analysis might yield some interesting suggestions on how to pick coauthors and paper length to minimize the review time. This could also help make the system go faster, no? Should we request these sort of statistics from the EiC?
Since I might be teaching detection and estimation next semester, I’ve been thinking a little bit about decision rules during my commute down the New Jersey Turnpike. The following question came to mind:
Suppose you see a car on the Turnpike who is clearly driving dangerously (weaving between cars, going 90+ MPH, tailgating an ambulance, and the like). You have to decide whether the car has New Jersey or New York plates [*]?
This is a hypothesis testing problem. I will assume for simplicity that New York drivers have cars with New York plates and New Jersey drivers have New Jersey plates [**]:
: New Jersey driver
: New York driver
Let be a binary variable indicating whether or not I observe dangerous driving behavior. Based on my entirely subjective experience, I would say the in terms of likelihoods,
so the maximum likelihood (ML) rule would suggest that the driver is from New York.
However, if I take into account my (also entirely subjective) priors on the fraction of drivers from New Jersey and New York, respectively, I would have to say
so the maximum a-posteriori probability (MAP) rule would suggest that the driver is from New Jersey.
Which is better?
[*] I am assuming North Jersey here, so Pennsylvania plates are negligible.
[**] This may be a questionable modeling assumption given suburban demographics.
Since I’m sick and I can’t really focus on math right now, here’s a flowchart to help you decide if you should go into campus.
Setting: a lone house stands on a Scottish moor. The fog is dense here. It is difficult to estimate where your foot will fall. A figure in a cloak stands in front of the door.
Figure: [rapping on the door, in a Highland accent] Knock knock!
Voice from inside: Who’s there?
Voice: Glivenko who?
[The fog along the moor converges uniformly on the house, enveloping it completely in a cumulus.]