The editor-in-chief has a new stick

Here’s a little bit of news which may have escaped notice by some in the information theory community:

“In view of its concerns about excessive reviewing delays in the IT Transactions, the BoG authorizes the EiC in his sole judgment to delay publication of papers by authors who are derelict in their reviewing duties.”

Reviewers may be considered to be derelict if they habitually decline or fail to respond to review requests, or accept but then drop out; or if they habitually submit perfunctory and/or excessively delayed reviews.

I’ve said it before, but I think that transparency is the thing which will make reviews more timely — what is the distribution of review times? What is the relationship between paper length and review length? Plots like these may shock people and also give a perspective on their own behavior. I bet some people who are “part of the problem” don’t even realize that they are part of the problem.

The card-cyclic to random shuffle

I heard about this cool-sounding seminar at the Berkeley Statistics department:

Mixing time of the card-cyclic to random shuffle
Ben Morris

We analyze the following method for shuffling n cards. First, remove card 1 (i.e., the card with label 1) and then re-insert it randomly into the deck. Then repeat with cards 2, 3,…, n. Call this a round. R. Pinsky showed, somewhat surprisingly, that the mixing time is greater than one round. We show that in fact the mixing time is on the order of \log n rounds.

The talk is based on a paper with Weiyang Ning (a student at UW) and Yuval Peres. The description the results is somewhat different because it’s \log n rounds of n moves, or n \log n moves. From the intro to the paper:

To prove the lower bound we introduce the concept of a barrier between two parts of the deck that moves along with the cards as the shuffling is performed. Then we show that the trajectory of this barrier can be well-approximated by a deterministic function f… we relate the mixing rate of the chain to the rate at which f converges to a constant.

The key is to use path coupling, a technique pioneered by Bubley and Dyer. It’s a cute result and would be a fun paper to read for a class project or something, I bet.

Five career-oriented things I wish I had done more of in grad school

As I’ve started my nascent academic career, I’ve faced a number of new and difficult challenges, and on more than one occasion I’ve felt hoisted with my own petard. Looking back on my graduate school experience, I realized there were a number of things which I could have done more that would have made things easier now.

  1. Write more grants. I came in to grad school with a NDSEG fellowship and then my advisor was well-funded. I had a blesséd life (especially compared to those in other departments). I should have gotten a bit more on the ball helping to write grants, even if it was just for the experience. Reading a proposal (or many) really gives a sense of what the document is about. More importantly, it’s a grant in your field and that’s something that you can’t really get out of one of those panels on grantsmanship that are pitched to a broad audience. Learning the process early has two benefits : it demystifies the strategy space in terms of writing, and it gives a good sense of what kind of things make sense and are “fundable.” I wrote little dribs and drabs of proposals here and there but it wasn’t until my postdoc that I really got to see and work on a full proposal.
  2. Do an internship. : It’s sad to say, but I never did an internship while in grad school. This was a real mistake. As an engineer it’s important to learn what is going on in “real” engineering but more importantly, as a engineering theorist it’s important to understand what’s going to be important in the future. Some of that you can do by watching the news and trends, but having a visceral sense of the challenges in making a real object is an important perspective. On the more practical side, it gives you more contacts in industry and the more people you know, the more ideas you can integrate.
  3. Constantly tell stories. : Part of this is for grant writing, but it’s a more general skill that is important for the job market and interacting with your colleagues, 99% of whom will be outside your area. This is a skill that is hard to develop in graduate school because we spend our time in our labs talking to other graduate students who are not that far away from us in terms of intellectual background. This kind of story-telling is often cast as “be able to talk to the general public” or “explain your research to a crowd that includes information theorists and device physicists.” Part of being able to tell a story about your research is being able to tell it in someone else’s language, which means understanding a bit about how people in other fields think. How would you explain your research to an anthropologist? I did practice explaining to other people, but I wish I had practiced more.
  4. Find a community. : I do a lot of work that falls between fields or on the corners of fields, so I’ve bounced around a bit. The last time I applied for jobs I had interviews wearing a networks hat, a signal processing hat, and an information theorist hat. Finding a community can have a positive or negative connotation. On the one hand, it’s dumb that people care so much about labels : “oh, he’s a machine learning guy,” or “oh, he’s a signal processing guy.” On the other hand, cultivating professional relationships with a research community is valuable because those are the people who will remember you when your name comes up.
  5. Write the journal version first. People always say this. I think Lalitha told me this in 2005 at my first conference. Of course it’s a good idea. And it’s a slippery slope if you let it slide for one paper… 5 years later you realize you have 4 nascent journal papers with nice full stories but not enough time to do the last bit of research or flesh out the results a bit more. Not every conference paper is (or should be) a journal-worthy idea. But the goal is to tell bigger stories than 4 page ICASSP paper, and it’s important to keep that bigger picture in mind. I wish I had done more work contextualizing some of the things I worked on so I could come back to them and expand on them later. Instead I have a lot of weird half-baked ideas in old PDFs sitting around on my hard drive.