How science publishing works

This is pretty much how it works, yes.

Business donuts are strictly ornamental

Last time I was down at U of T, their career center had a book posted in its window, entitled Don't take the last donut: New rules of business etiquette.

Let's take the title advice seriously: as a matter of business etiquette, don't take the last donut. Why? Obviously, because it's impolite to deprive someone else of the option to have a donut.

Now assume that everyone is polite, and suppose there are two donuts left. What if you take the second-to-last donut? It's impolite for anyone else to take the last donut, so since everyone is polite (by assumption), your action deprives everyone else of a donut. Thus, it must be as impolite to take the second-to-last donut as it is to take the last one. We can make a similar case for the third-to-last and fourth-to-last donut.

Indeed, by the same argument, we can construct an inductive case: assuming it is impolite to take the nth-to-last donut (for any n >= 1), and there are (n+1) donuts available, then it must be equally impolite to take the (n+1)th-to-last donut, because nobody can (politely) take a subsequent donut.

Thus, by induction, for any integer n > 0, it is impolite to take the nth-to-last donut. Thus, business donuts are strictly ornamental and it is impolite to ever eat them. QED.

PLoS One: Publish in good journals, reject more papers

A neat article in PLoS One about peer review: it turns out that your rejection rate as a reviewer is related more to where you publish than your experience as a researcher. And if you publish in higher-quality journals, you're likely to reject more papers.

There are lots of ways to analyze and criticize this result, but here's my take. If you generally publish in higher quality journals, you will be doing most of your reviewing for higher-quality journals, which are more selective.

As I mentioned, the authors didn't find a relationship between experience and rejection rate, but I thought this figure was interesting: early researchers are clustered around the same "rejection intensity", while later ones tend to be spread out between much more stringent and much easier.

Shameless self promotion: Transactions on Info Theory Edition

My recent paper in the Transactions:

A. W. Eckford, “Ordering finite-state Markov channels by mutual information,” IEEE Trans. Inform. Theory, vol. 55, no. 7, pp. 3081-3086, Jul. 2009. [PDF]

In a Markov channel, the channel parameter is selected by the state of a hidden Markov chain -- the most famous example is the Gilbert-Elliott channel, in which a channel use might see a "good" state with low crossover probability, or a "bad" state with high crossover probability.

In Gaussian noise, channels are ordered with respect to SNR -- we know that a channel with smaller SNR has smaller capacity. In this paper, I'm trying to find a similar ordering for Markov channels: the "degraded family" of a "parent" channel, where all channels in the family have smaller capacity than the parent.

This week in charcuterie

Bacon-flavored bacon gum.

Strawberry-flavored gummy bacon.

Edible Rubik's Cubewich.

[via Gawker/Gizmodo.]

Nortel in liquidation

Here's what astonishes me. The federal government would drop 7 billion dollars to prop up a badly run company that has done its best to make itself obsolete. This same government then barely lifts a finger as foreign interests acquire the key assets of another badly run company that is at the forefront of the next generation of wireless networks.

Since the bidders were most likely interested in Nortel's patent portfolio, all those high-tech R&D jobs are either going or gone. This, and not the troubles in the auto sector, is the story that will have lasting ramifications for the Canadian economy.

Hot topics in info theory

I'm trying to figure out what's hot and what's not in information theory. So, for a laugh, I put together all the session titles from ISITs 2005-2009, and searched them for keywords.

In the following, I'm giving counts of the number of times a keyword appears in the list of sessions. The numbers of sessions at each conference were slightly different (from a low of 126 in 2005 to a high of 152 in 2006), but I don't bother to normalize since the range is small. Some findings:

MIMO is out, wireless is in:

Number of sessions with the given keyword:

Year            | 2005   2006   2007   2008   2009
----------------|----------------------------------
MIMO            |   5     14      7      6      0
wireless        |   3      5      5      4     14

Relay is on the downswing, cooperation is picking up the slack:

Number of sessions with the given keyword:

Year            | 2005   2006   2007   2008   2009
----------------|----------------------------------
relay           |   4      2      9      8      5
cooperation*    |   1      3      3      4      4

* includes "cooperative"

Turbo vs. LDPC -- LDPC is resurgent:

Number of sessions with the given keyword:

Year            | 2005   2006   2007   2008   2009
----------------|----------------------------------
turbo           |   3      0      1      1      1
LDPC            |  11     16      6      6     11

Source vs. channel -- sources are on the decline:

Number of sessions with the given keyword:

Year            | 2005   2006   2007   2008   2009
----------------|----------------------------------
source          |   6     15     11      9      6
entropy         |   1      3      1      1      0
                |
channel         |  12     19     25     16     22
capacity        |   7      7     13     11      9

I've posted the session lists here if you want to do your own grepping: 2005, 2006, 2007, 2008, 2009.