Andrew Eckford: The Blog

Another reason to delete conference spam

2012-01-31T19:33:00.003-05:00

If you're like me, you get an immense amount of conference spam, even for conferences that are not remotely in your area of interest. (This morning I got a message inviting me to an international geography and geology conference. All right then.)

According to a new report, conference invites are now being used to infect victims' computers with malware. It works like this: the attacker takes the PDF Call for Papers for a legitimate conference, infects the PDF with malware, and spams it to various targets as a conference invitation (with the malicious PDF as an attachment). The targets open the PDF and get infected.

So far, the affected conferences include ISSNIP, an IEEE-sponsored conference on sensor networks; but the malware version of the CfP may have only targeted one particular defense contractor.

(Previously in our irregular series on academic spam: 1, 2)

BIRS ... brrrrr

2012-01-17T16:42:00.000-05:00

This week I'm at a workshop on Interactive Information Theory at the Banff International Research Station (BIRS). More or less, interactive information theory is what happens when transmitter and receiver are allowed to go back and forth. This includes things like feedback capacity, interactive function computation, secret key exchange, and so on.

It's been an outstanding workshop so far and I'll blog more about it later. But it's been very cold. (Not to complain. I grew up in Edmonton so this is like coming home for me.)

We just took the group photo, below, so you can play "name that information theorist". (I'm in the yellow parka and grey hat on the right hand side.) Weather conditions when the photo was taken: light snow, -30C.

An odds start for the new year

2012-01-02T11:46:00.000-05:00

What's going to happen in 2012? I like odds better than predictions, because future events are inherently uncertain. Here are the odds I would give on events in 2012, if I were a betting man. Add yours in the comments.

(Odds are of the form against-for, e.g. 10-1 is 10 to 1 against, 1-10 is 10 to 1 for. Rely on them at your own risk.)

Canadian Research and Universities

NSERC budget cut: 1-2
NSERC budget increased more than 5%: 10-1
Canadian wins Nobel prize: 5-1

Science and Technology

P=NP? solved: 1000000-1
Faster-than-light neutrino result debunked: 1-10
Higgs boson found: 1-20
D-Wave One is accepted by Aaronson as a quantum computer: 10000-1
DIDO is implemented in a commercial product: 100-1

... and has the claimed performance: 100000-1

York University

CUPE 3903 strike: 4-1
YUFA strike: 8-1

Canadian Wireless Industry

RIM acquired: 1-1

RIM acquired by Microsoft: 2-1
RIM acquired by Nokia: 5-1
RIM acquired by Amazon: 10-1

RIM shares end year (or sold for) above $20: 1-1

... or below $10: 2-1

Both Mobilicity and Public Mobile acquired: 1-5

Mobilicity acquired by Globalive: 1-1
Public Mobile acquired by Bell or Rogers: 2-1

Two of the big three (Telus, Bell, Rogers) announce merger: 10-1

... and the merger is permitted by regulators: 20-1

Toronto and Canada

Leafs make the playoffs: 2-1
Leafs win the Stanley Cup: 100-1
Next leader of the NDP is Brian Topp 1-1

... Thomas Mulcair: 3-1

Next leader of the Liberal party is Bob Rae 2-1

... Justin Trudeau 5-1

End-of-year roundup

2011-12-19T10:56:00.004-05:00

A few end-of-year items:

I was in York (that's York, UK) a couple of weeks ago to attend Bionetics. I presented the following paper, in which we move beyond analysis of active transport molecular communication, and get into channel design:

N. Farsad, A. W. Eckford, and S. Hiyama, "Channel design and optimization of active transport molecular communication," in Proc. 6th International ICST Conference on Bio-Inspired Models of Network, Information, and Computing Systems, York, UK, 2011. [PDF]

Early in the new year I'll be attending this workshop at BIRS. It looks even more tempting than the January skiing in Banff, and I'll try to blog about it if I find the time.

The EE program proposal, part of York's rapid engineering expansion, is grinding away -- it's about to go up for approval by the School of Engineering. Meanwhile, here's our first job ad. We're focusing on electronics, power, and medical devices in this round, but we plan to add signal processing and communications spots later (and maybe now, if we get a truly outstanding candidate). Tell your friends!

Happy holidays / Joyeux fêtes, and I'll see you in 2012.

Weird Sun Media - York U Internet Problem (Update 2)

2011-11-30T10:21:00.001-05:00

Are you at York University right now? Quick, try to connect to torontosun.com, ottawasun.com, winnipegsun.com, calgarysun.com, and edmontonsun.com.

Add a comment below (or email aeckford@gmail.com if you don't want it to be public) with the following info:

What do you see?
What's your computer, operating system, and browser?
What's your IP address? (fastest way to find it: use Google)

I'll update when I find out more.

Update Dec. 3: The outage is fixed! I got a call from Sun Media's Director of Networks, Telecom, and Information Security. He asked me to send some technical details, and their tech team seems to have solved the problem. It's not clear what happened, but the director mentioned that it might have been in response to hacking attempts, as I suspected. (Meanwhile, the only response I got from York's UIT was to close my support ticket without explanation.)

Update Dec. 1: The outage is real. I can't reach the Sun papers from either of the computers in my office (in the cse.yorku.ca subnet), nor can I reach them when I connect to AirYork. The outage has lasted several days at least. However, the outage is not universal to Sun Media sites (listed here). Only the above five sites are affected.

To be clear, I'm not in my office all day reading Sun newspapers on the web. I'm not exactly in their target demographic. But I think we should all be concerned when there are barriers to information access at a university. More details after the jump.

Interestingly, all five sites resolve to the same IP address: 207.253.226.30 (and as far as I can tell, those are the only sites that map to this IP address). IP addresses are normally associated with individual servers (or groups of linked servers), so the problem likely has to do with handling of requests by that server.

I tried traceroute, which at least established that York is not blocking outbound packets. There are two other possibilities, one that York is blocking inbound packets from 207.253.226.30, or that 207.253.226.30 is blocking inbound packets from York.

I think the latter explanation is more likely. If I were to speculate, perhaps 207.253.226.30 has detected hacking attempts originating in the York subnet, and blocked access in response (maybe automatically).

I tried to contact York's UIT, but they haven't returned my message. Next step is to contact Sun Media.

If I were Dean of Graduate Studies: Rethinking the PhD

2011-11-22T09:13:00.001-05:00

With the departure of Douglas Peers to Waterloo, York University is now looking for a new Dean of Graduate Studies. So let's say I put my resume in, and let's also say all the other candidates got struck by lightning. What would I do if I were Dean?

I've sat on the graduate faculty council, so I'm well aware that the nuts and bolts of graduate administration is mundane; even the controversies are kind of boring. But as your new Dean of Graduate Studies, I would address an issue that I don't see discussed very much: what should a graduate faculty be teaching?

My first principle would be this: education should be useful in its own right, not necessarily as a means toward a particular job; this logic should apply to any degree that a university awards. In the case of the PhD, the faculty of graduate studies need not be an intellectual trade school, in which students learn to be professors. In any case, the model of the PhD as an apprenticeship for the tenure track is broken, and it is time that graduate faculties acknowledged this fact.

Instead, I would rework the graduate curriculum to teach a skill that is useful in itself, and something that every PhD should learn (but that not all do): how to think originally.

In suggesting this, I'm arguing that original thinking is a skill that can be learned. The best graduate students I've seen in my career, both under my own supervision and others, have had a skill set that includes the following abilities:

Recognition that a problem is original, and finding related work. Being able to find existing solutions in the literature is key, and often is only possible by thinking about the problem in many different ways. The first few keywords you type into Google won't find exactly what you're looking for.
Knowing how to get started on a previously unsolved problem. Here, skills like abstraction, modeling, constructing toy examples, and thought experiments may be useful.
Knowing how to catch errors. If a problem is unsolved, then the solution is by definition unknown -- so how can we rely on such a solution? There are lots of ways to check, like sanity checking (making sure the solution follows basic rules, like energy can't be negative, probability has to be less than 1, and so on).
Running with an idea. Original ideas naturally lead to more problems. Once the first problem is solved, what's the next logical step? And the next one?

Surely it must be possible to create a graduate curriculum that teaches students how to do this, regardless of the specialty (at least in the sciences). And right now, graduate faculties do a surprisingly inconsistent job at teaching these skills: it is more or less up to the graduate supervisor to ensure that students learn them.

I'm not talking about teaching students how to be geniuses; I'm talking about teaching the nuts and bolts of creating new knowledge. If the PhD does not teach these skills (formally), then what should it teach?

UPDATED: Big news for York Engineering

2011-11-01T10:15:00.003-04:00

UPDATE Nov. 2: York science librarian John Dupuis livetweeted the event (and collected a summary of tweets here). From John's blog:

I did a ton of live tweeting of the announcement, some of which is here, here, here, here, here, here, here, here, here, here, here, here, here, here, here, here, here and here. The most significant new information revolves around naming the new engineering school The Lassonde School of Engineering. As well, what is now the Computer Science and Engineering Building will be known as the Lassonde Building.

Original post follows.

Canadian entrepreneur Pierre Lassonde is donating $25 million to expand engineering programs at York University. The donation will be matched by the university, and combined with the earlier announcement of $50 million for a new engineering building, that makes a total of $100 million for engineering expansion at York. We've been told that Lassonde's contribution is the largest single private donation in university history.

The official announcement will happen today at 1 PM, but in the meanwhile you can read more about the donation in this Globe and Mail story.

With the donation, the university's goal is to move away from the "niche" engineering programs currently in the program, and become a full-fledged "traditional" engineering school. You can expect York to add traditional engineering disciplines like civil, mechanical, and chemical over the next few years.

However, it's electrical engineering that will take the lead in expansion. I am chairing the committee that will write the EE proposal, and we're operating under the assumption that the first students will be admitted in 2013. I'll try to blog more about our progress as things develop.

For now, it's definitely an exciting time to be at York.

UPDATED: Talk at Case: Models and Capacities of Molecular Communication

2011-10-26T09:33:00.000-04:00

Updated Oct 26, 2011: Here's the video of the talk (this is a playlist). Try HD (720p) and fullscreen, which makes it easier to read the slides.

Original post from October 13 follows.

I'm continuing my 2011 "lecture tour" with a visit to Case Western Reserve University in Cleveland, Ohio. My host is Prof. Peter Thomas, and he would be able to give you further details.

"Models and Capacities of Molecular Communication"

Friday, October 14, 11:30 AM

Abstract: What are the fundamental limits of diffusion-mediated molecular communication? This question, which has only recently attracted attention from information theorists, turns out to be surprisingly difficult. Not only is the communication medium unfamiliar to communication engineers; but the mathematical details of the communication environment are complicated. In this talk, we discuss mathematical models for molecular communication, which are both information-theoretically useful and physically meaningful; we discuss the difficulties of dealing exactly with these models; and we present some simplified scenarios in which capacity can be evaluated. Finally, we discuss the engineering and biological significance of these results.

Fixing undergraduate education

2011-10-24T10:38:00.002-04:00

Here's Jeffrey Simpson's column this past Friday -- his latest in a series (1, 2) taking universities to task over undergraduate education. (The Globe seems to be on a tear lately about our "unsustainable" universities.)

I seem to live in a different world than Simpson's hard-luck undergrads: in my program, classes are small and are (almost) all taught by full-time, tenure-track faculty. The small classes probably come from the newness of our program -- we're expecting to ramp up enrollments over the next decade. But the use of full-time faculty in teaching is more or less universal at Canadian engineering schools. Why is that? It's partly because engineers with PhDs can get lucrative jobs, so perma-adjuncting is not very attractive. But it's also because you need a license to teach engineering in Canada -- licenses are tough to get, and until very recently you couldn't get one if teaching was your only work experience.

So here's something the provincial government could do, which would cost very little up front: require teachers at public universities to be licensed, much as public school teachers are. As a condition of the license, the province could require a course or two on teaching at the postsecondary level (something very few professors have ever received -- right now it's a learn-on-the-job kind of situation). The province could also distinguish between "adjunct" licenses and "tenure-track" licenses, and require at least some minimum fraction of education hours be given by tenure-track license holders.

This kind of solution strikes me as much easier to implement than trying to strong-arm universities into opening up collective agreements with faculty, as Simpson seems to want. It wouldn't do much about class sizes, but that is more of a monetary issue.

Of course, I disagree fundamentally with Simpson's view of the university -- particularly his disregard for research and graduate education -- but that's a topic for a different post.

Stimulus, Response: The G&M on Universities in Canada

2011-10-13T13:50:00.000-04:00

On Tuesday, the Globe and Mail called out Canadian universities for their poor undergraduate teaching, calling current practices "unacceptable" and "unsustainable".

Here's one of many twitter discussions it triggered, letters to the editor (scroll to "Connecting the Dots"), and blog-length responses by Melonie Fullick, OCUFA, and Worthwhile Canadian Initiative.

There's lots of good stuff to read there, and I only have this to add: the Globe editorial frames the problem in terms of tension between research and teaching, and disregards any role for the modern university beyond the training of undergraduates. This is consistent with positions expressed by their senior columnists, Jeffrey Simpson and, more outrageously, Margaret Wente (seriously, go read Wente's column if you haven't already).

The "tension" between teaching and research is about as real and helpful as the War on the Car. Strong research programs help faculty keep their teaching dynamic, current, and topical; besides which, graduate-level teaching (like supervision of a Ph.D.) is research, and is totallly ignored by the Globe. So let's please put to bed this corrosive idea that research is some kind of luxury, or an optional activity that can be cut without consequences. (Again, where have we heard this kind of simplistic thinking before?)

[h/t @qui_oui for the links]

Talk in Kobe: Information-theoretic problems in molecular and nanoscale communication

2011-09-24T18:48:00.004-04:00

I'm in Osaka right now on a research trip, getting caught up on my many Japanese research contacts. One of the items on my agenda is a talk in Kobe on Monday afternoon, at the Kobe Advanced ICT Research Center, part of NICT. Details:

Title: Information-theoretic problems in molecular and nanoscale communication

Abstract: Recent advances in MEMS/NEMS and systems biology have made it possible to manufacture customized nanoscale devices, such as swarms of nanorobots. However, manufacturing is not enough: a significant remaining challenge is to solve the communication problem among these devices, which would allow them to coordinate their actions.Furthermore, the nanoscale communication environment is rather different from the systems that are usually studied by information and communication theorists. In this talk, I will introduce the nanoscale communication problem from an information-theoretic perspective, focusing on molecular communication, which mimics the way in which microorganisms communicate.Mathematical models and achievable-rate results will be presented, and important open problemswill be discussed.

[Link for further details]

The ICC Deadline Fiasco

2011-09-20T10:41:00.001-04:00

ICC is one of ComSoc's "flagship" conferences, and I try to submit something every year. It is more or less an annual tradition that the submission deadline is extended at least once. And although extensions are widely expected, they are never certain -- the conference organizers use the magic words (see here):

The Conference TPC has set realistic deadline dates and these may be rigorously adhered to.

This year, the original deadline was on Tuesday, September 6. Late in the evening of September 6, the first extension was officially announced by email, to September 19. And this morning, after working late into the night to get our paper finished, I noticed (not by email -- there has been no official announcement as yet -- but by logging in to EDAS) that the deadline has again been extended to September 28.

To be fair, the extensions have been useful. We were having trouble getting good results out of our simulations, and without the first extension, we wouldn't have had a paper. But it is really annoying that the extensions were announced with only a few hours to go before the deadline. In the most recent case, I spent the weekend working late, and cleared my Monday schedule to make extra time to hit the deadline. For people like me with families and busy jobs, that causes a lot of tension.

I've never served any high-up role on a large TPC, so I don't know how extension decisions are made. But I don't think it's too much to ask that the extensions be announced with 48 hours' notice. It's not nice to let people ruin their weekends for no reason.

UPDATED: Paper at PIMRC: Hardware implementation of fractional cooperation

2011-09-12T13:47:00.001-04:00

Update Sept. 15, 2011: PDF of the paper is here. Citation:

A. Calce, N. Farsad, and A. W. Eckford, “An experimental study of fractional cooperation in wireless mesh networks,” in Proc. 22nd Annual IEEE Symposium on Personal Indoor Mobile Radio Communications (PIMRC), Toronto, ON, pp. 990-994, 2011.

Original post follows.

I've got a paper at this week's IEEE PIMRC in Toronto:

A. Calce, N. Farsad, and A. W. Eckford
An experimental study of fractional cooperation in wireless mesh networks
Session: LPAN-8 (Multihop Cooperative Communication)
Date: Tuesday, September 13, 10:30-12:00
Room: Pier 9

Over the past five years I (and a bunch of other researchers, like Ravi and Nariman) have done a pile of work on fractional cooperation. Under this scheme, cooperation partners can just show up, relay as much of your transmission as they feel like (selecting symbols from your frame randomly), and leave -- with very little protocol and essentially no coordination. As chaotic as this sounds, we have always been able to show good results in simulation.

But with this paper, we put fractional cooperation in hardware for the first time: using some spare iMotes we had in the lab, my student Anthony managed to put together a small mesh network, showing that the system works as well in practice as it always has in MATLAB.

I'll post PDFs soon. Anthony will be giving the talk.

Designing microchannel molecular communication systems: Paper in Nano Communication Networks

2011-08-29T09:31:00.009-04:00

I've got a new paper in Nano Communication Networks. (Of the five journal papers I've co-written on molecular communication, this is the last to be submitted and the first to be published ... quick review process FTW. But on the other hand, it's published by Elsevier.)

N. Farsad, A. W. Eckford, S. Hiyama, and Y. Moritani, “Quick system design of vesicle-based active transport molecular communication by using a simple transport model,” Nano Communication Networks, doi:10.1016/j.nancom.2011.07.003, 2011. [PDF]

In this paper, we're continuing our work on microchannel molecular communication (see also here), where molecular motors are used to transport the message-bearing molecules (MBMs).

These systems work more or less as follows. MBMs are arranged on a "loading zone" at one end of a microchannel. Every so often, a motor arrives at the loading zone, collects a random number of MBMs (possibly zero), and transports them to the other end of the microchannel, dropping them off at an "unloading zone".

The receiver's job is to figure out the message based on whatever arrives at the unloading zone. For instance, say we want to transmit a binary message, 0 or 1. If we want to send 0, we place 0 MBMs on the loading zone, and if we want to transmit 1, we place 1 MBM on the loading zone. If 0 is sent, then obviously 0 MBMs arrive at the unloading zone, because there are none to start with. On the other hand, if 1 is sent, then either 0 or 1 MBMs arrive, because the motors might not pick up the MBM available at the loading zone. Thus, we have a noisy communication channel -- in this example, it is equivalent to the Z channel.

It turns out that these systems are very complicated to simulate, especially since the probability of each MBM arriving at the unloading zone is dependent on its neighbors. In this paper, we give a low-complexity design tool to speed up simulations, and use it to design optimal loading zones, which maximize the probability that the motors will pick up MBMs.

Generating Lévy random variables from Gaussian (Updated)

2011-08-17T11:08:00.000-04:00

Update August 17, 2011: Corrected a typo in the distribution.

Sometimes when I learn a neat mathematical trick, I write a blog post so I don't forget it. This is one of those times.

If a random variable x has the Lévy distribution with parameters c and

μ, then the pdf of x is given by

This distribution has applications in economics, finance, and physics. A famous statistical application is the first passage time of a Brownian motion: if w(t) is given by the Wiener process (with initial condition w(0) = 0), and x is the first time that w(x) = d, then x is

Lévy distributed with c = d^2 and

μ = 0.

Here's the trick, which is hard to find in the literature: let z represent a Gaussian-distributed random variable with mean 0 and variance v. Then 1/z^2 is distributed

Lévy with c = 1/(2v^2) and

μ = 0. In other words, if you have a good Gaussian random number generator, you can use it to quickly generate

Lévy-distributed random variables!

This property is very briefly mentioned in the following paper (I had to do a bit more digging to verify it and get the right parameter values):

J. M. Chambers, C. L. Mallows, and B. W. Stuck, "A method for simulating stable random variables," J. Am. Stat. Soc., vol. 71, no. 354, pp. 340-344, Jun. 1976.

Talking about wireless: Elizabeth May's advice

2011-08-08T09:50:00.002-04:00

Green Party leader and Member of Parliament Elizabeth May recently caused a stir by expressing concern over the safety of electromagnetic radiation. And here she is again this past weekend, taking an interview with her BlackBerry set to speakerphone, to avoid radiation exposure.

I'm not going to go into her concerns about safety (with which I largely disagree; if you're interested in the debate, an excellent summary is here). At the very least, May makes the valid point that there is no scientific consensus on wireless safety. Fair enough!

But in her longer critique of wireless safety (found here on her blog), and by way of her subsequent actions, May gave some advice to those who are concerned about their exposure to EM radiation. Paraphrasing, she said this:

When using your phone, use speakerphone or text rather than holding it up to your ear.
You can use a smartphone, but don't keep it in your pocket.
At home, use a wireline network, not WiFi.
Don't get a Smart Meter.

What I will talk about is this: for people who are concerned about radio exposure (which doesn't include me), did Elizabeth May give good advice? It seems to me that #1 is good advice, #2 and #3 are questionable, and #4 is wrong to the point of embarrassing.

Let's get this out of the way first: since nobody knows how wireless signals might cause cancer (even assuming they do at all), it's impossible to say for sure. Even figuring out average exposure is not easy without something like a portable spectrum analyzer, which I don't have.

But in true engineering tradition, I'm going to pick some numbers and try to come up with a back-of-the-envelope answer. I'll make the following assumptions:

Wireless signals obey the inverse square law: for distance d, power is proportional to 1/d^2. This assumption is not always accurate, but it's pretty good under most circumstances. Indoors, it is usually a worst-case assumption -- it will often overestimate your exposure.
All wireless devices transmit with the same power: This assumption is totally inaccurate -- power varies not only by device, but also by type of use (phone, text, browsing, etc.) -- but it's easy to work with. And it's again a worst-case assumption if you're comparing other devices to cell phones, since cell phones generate more radio power than almost any other consumer device. (I'll also make some comments about relaxing this assumption.)

May's first suggestion -- text or use speakerphone rather than hold the phone up to your ear -- is a good one, if you're worried about exposure to radiation. And she seems to be comfortable doing this herself (e.g., in the interview with the Globe), so let's use this as a baseline: on speaker or text, I'll usually hold the phone about 50 cm away from my head, so that gives us a starting point for power comparisons.

How about keeping the phone in a pocket? For one thing, there's nothing but fat, muscle, and skin tissue in your hip, but the links to cancer all focus on the brain -- no worries about cancer anywhere else from any literature I'm aware of. At about 1 m away from the head, a phone in a pant pocket is sending even less "harmful" radiation to the head than on speakerphone: about a quarter the power, since 1 m is 2 times 50 cm, and 2^2 = 4. And if May doesn't keep it in her pocket, where does she keep it? A briefcase would be better, but if she keeps it in a shoulder bag, it might be even closer to her head than if it was in her pocket. So this is kind of questionable.

What about WiFi? I sit about 1 m away from my computer, and as I type this I'm about 5 m away from my WiFi router. Compared to texting, my computer is about twice the distance away (2^2 = 4, so a factor of 4 weaker); and the router is about 10 times the distance (10^2 = 100, so a factor of 100 weaker). On top of that, WiFi signals are normally much weaker than cell phone signals -- because WiFi is supposed to cover just your house, while cell phones are supposed to cover your whole neighborhood. But even using our worst-case assumptions, and even assuming you stay 5 m away from it all day and don't go outside or anywhere else in your house, your WiFi router is giving you a daily exposure less than that of a 15 minute cell phone call on speakerphone. So if May is okay with using speakerphone, it's odd to me that she would not want WiFi installed in her house.

Finally, let's talk about Smart Meters. These devices use the same family of frequencies as your WiFi router (called the ISM band), and by law their wireless power emissions are constrained: again, worst case, they use about the same power as a cell phone. What's more, the are normally installed outside your house, on the other side of a wall (which itself absorbs 1/2 to 3/4 of the power). And what's more than that, the amount of information they have to send is tiny -- your electricity use doesn't change that fast! -- so they are only active, on average, about 0.1% of the time. So let's consider the worst possible case: the meter is located on the wall directly opposite where you sleep, and your head is more or less right against the wall, about 50 cm away from the meter. Then over your 8-hour sleep, counting the meter's transmitted power, the absorption of the wall, and the meter's low transmission rate, you are absorbing at most the equivalent of a fifteen second cell phone call on speakerphone.

Considering the kind of RF exposure Ms. May finds acceptable, and the kinds of energy conservation benefits that smart meters make possible, this stand against smart meters lacks perspective and is just plain embarrassing. It's the radio frequency equivalent of eating a giant ice cream sundae, then ordering a diet soda because you're trying to cut down on sugar.

First Shannon Award Livetweet

2011-08-04T13:41:00.000-04:00

Announced moments ago at the ISIT 2011 banquet in St. Petersburg, Russia: Congratulations to Abbas El Gamal on his well-deserved Shannon Award.

Livetweeted from the event by Johann Briffa:

And now a word from our sponsors (repost)

2011-07-26T09:02:00.003-04:00

(This post was originally published in March, but was taken down as -- unknown to me at the time -- the results of the Discovery competition were under embargo. Now that the results are public, I'm republishing the post.)

NSERC Discovery results came out yesterday. This is the program that supports almost all of the curiosity-driven, non-industrial, basic science and engineering research in the country. It's generally not a huge amount of money -- enough to pay 1-2 graduate students is typical. But most Canadian professors hold one, and it probably pays the salary of the majority of Canada's science and engineering graduate students.

Just now, a summary of the competition landed in my inbox. By now, the "new regime" of Discovery funding is well known: applications are assigned a quality score, and a pot of money is assigned to each score value, divided among the applications with that value; below some score the amount is zero. Applications are now "memoryless", meaning that the status and funding level of your last application have no bearing on your current application (I would argue this is bad for all kinds of reasons, but that's another discussion. See Ghoussoub's excellent blog for detailed summaries and a discussion of what's going on with NSERC.)

But the following details were interesting:

For the 2011 DG competition, the EGs reviewed 3,482 individual and team applications. The results, by category of applicants, are as follows:
The percentage of early career researchers receiving a first grant is 54 percent, surpassing NSERC’s target of 50 percent. The average grant is $22,481.
The success rate for established researchers renewing a grant is 74 percent. The average grant is $35,045.
The success rate for established researchers who did not hold a grant in 2010 is 33 percent, with an average grant of $28,082.

I had an application in this round -- I won't say what I got, but I will note that I did get a grant (huge relief!).

Target or no, 54% for first-time applicants is remarkably small. Discovery is one of the few programs that is realistic for an early-career researcher; most of the others require industrial collaborators, which are hard for new professors to cultivate. Further, first-time applicants are facing a ticking tenure clock, and need to get graduate students in the door as quickly as possible -- and they need money to do that.

These numbers also indicate that the amount of the average first-time grant is remarkably high; generally more than one would need to employ a single student. To me, it would make more sense to spread the money around more evenly, to give early-stage researchers a chance to make their careers.

Also remarkably low is the renewal rate. Of the 26% who didn't get a renewal (comprising, I would estimate, hundreds of researchers), I would bet there are plenty who now have no way to fund their graduate students.

Let's get rid of Transactions letters (Updated)

2011-07-21T13:33:00.005-04:00

Update August 11, 2011: According to an editorial in the August issue of Transactions on Wireless Communications (not yet available on IEEExplore), both Trans. Comm. and Trans. Wireless Comm. will stop accepting letter submissions as of September 1, 2011. Original post follows.

Both the Transactions on Communications and Wireless Communications accept "letters". There are minor differences between the two journals, but letters are most commonly used as "enhancement of previously published work", i.e., minor results that fall short of a full paper. In both cases there are stringent length requirements: 10 double-spaced pages in the Transactions draft format excluding figures, which works out to about 4 pages in the two-column format.

Anyone who has written a journal paper -- or even a conference paper -- knows that 10 double-spaced pages is barely enough to say anything. As both a reviewer and an author, I've seen the following dynamic happen over and over:

Author discovers an interesting little result. Not being quite enough for a paper, author writes a letter, leaving out details that s/he considers irrelevant to the overall point of the work, in order to meet the length requirement.
Reviewers are not satisfied with the level of detail in the paper. Ignoring scope and length requirements for letters, reviewers demand more detail, additional simulations, longer explanations, and so on.
Author adds the absolute minimum the reviewers demand, and then cuts, edits, shortens, and otherwise takes a butcher knife to the paper to meet the length requirement.
Go to 2.

"Letters" end up in a kind of limbo, delaying the review process and making nobody happy, even if the paper is eventually accepted.

The problem seems to be that reviewers treat "letters" like regular papers. So why not make it official and drop the "letter" from the Transactions? Short papers could be handled just like regular papers, with the understanding that the magnitude of the paper's contribution should be proportional to its length. The Transactions on Information Theory did this years ago.

Papers at CWIT: PDFs and video

2011-06-24T09:23:00.001-04:00

I posted earlier about my papers at CWIT. I finally got around to uploading the PDFs and video.

J. P. K. Chu, A. W. Eckford, and R. S. Adve, “Distributed optimization of the Bhattacharyya parameter in wireless relay networks,” in Proc. 12th Canadian Workshop on Information Theory, Kelowna, BC, 2011. [PDF] (Video not available.)

L. Cui and A. W. Eckford, “The delay selector channel: Definition and capacity bounds,” in Proc. 12th Canadian Workshop on Information Theory, Kelowna, BC, 2011. [PDF]

Presentation video for the Cui/Eckford paper:

Jeffrey Simpson, do your research

2011-06-20T12:55:00.004-04:00

Good news, everyone! Jeffrey Simpson has solved all that ails the university sector in Ontario: it turns out that those lazy professors need to stop doing their airy-fairy research and get their butts in front of the classroom. Oh, and we also need heavy-handed government intervention to make this happen.

Simpson's column betrays an almost complete lack of understanding of the modern university, not to mention the role and importance of research. I would invite Simpson to check the research page of any university in Ontario, listing high-profile and news-making results (e.g., U of T, York). Perhaps in his next column, Simpson can let us know which of these results he would trade against smaller class sizes for undergraduates, from the innovative and censorship-fighting Citizen Lab to the breakthroughs of the University Health Network to environment-saving advances in clean energy. Or perhaps he can let us know which of the many innovative spin-off companies arising from university research Ontario can do without (at U of W alone, valued at nearly $1 billion annually in a 2001 report).

Simpson also fails to understand that there is no tension between research and teaching. Indeed, at the graduate level, research is teaching: graduate students learn the difficult process of developing new knowledge through research, under the close guidance of a supervising professor. From personal experience, I can say that graduate supervision is a rewarding but time-consuming process; one that Simpson completely ignores.

I could go on, but I've got a lot of research work to do this afternoon! I'm working on an innovative communication technique for nanorobots that may one day revolutionize drug delivery and make invasive surgery obsolete -- take that, taxpayers! Meanwhile, before writing his next column, I suggest Simpson do a bit of the research he hates so much and acquaint himself with what universities actually do in 2011.

On Slepian-Wolf compression, in memory of Jack Wolf

2011-06-08T09:32:00.007-04:00

Jack Keil Wolf, one of the giants of information theory, passed away last month. His obituary was carried in the New York Times.

Wolf's signature contribution was in distributed data compression. Say you're trying to retrieve two files, X and Y, from separate servers on a network. One strategy would be to compress both files individually to their entropy, H(X) and H(Y), and download them; this requires H(X)+H(Y) bits. This is as good as you can do if X and Y are statistically independent. However, if they are dependent, compressing X and Y together can be done at the joint entropy H(X,Y), which is less than H(X)+H(Y). But what can be done by the two servers, since neither has access to both X and Y?

Along with David Slepian, Wolf proved the Slepian-Wolf theorem[1]: as long as certain conditions are met, files X and Y can be compressed to H(X,Y), even if the X server has no knowledge of file Y, and vice versa. One way to do this is to have the X server just compress and transmit X at a rate of H(X), and have the Y server form the "difference signal" and transmit it at a rate of H(Y|X), for a total rate of H(X)+H(Y|X)=H(X,Y), called a "corner point" solution.

How does the Y server form a difference signal if it doesn't know X? The server groups all possible Y values into "bins", as follows: if X is known, then only one member of the bin can possibly be equal to Y. The server then sends some string or number which identifies the bin. Assuming the strategy is agreed upon in advance, the receiver sees H(X) and decodes X, then it sees the bin identifier from the Y server, and picks Y as the unique member of the bin corresponding to X.

For instance, suppose X and Y are temperature readings, and suppose we know from past experience that either Y=X, or Y=X+1 (because the sensors are not too far apart, say). Now, suppose X is even. Then if Y is even, it must be true that Y=X, and if Y is odd, it must be true that Y=X+1; and vice versa if X is odd. So if we know X, and we know whether Y is even or odd, then we know Y. Our strategy is then the following: server X encodes the temperature independently, and server Y sends one bit: 0 if Y is even, and 1 if Y is odd. Since a 2-digit temperature would normally require 7 or 8 bits to encode, this is a huge savings of data compared to encoding X and Y independently.

Years ago, I did a bit of work on Slepian-Wolf, when the big idea was to use the syndromes of a powerful error-correcting code (e.g., an LDPC code) to form the difference signal [2,3], or to form more general balances away from the "corner point" (what I did in [4]). I haven't worked in the field in years, but as far as I know, nonbinary Slepian-Wolf compression is still a major open problem, since LDPC codes have a hard time handling it.

[1] D. Slepian and J. K. Wolf, "Noiseless coding of correlated information sources," IEEE Trans. Inform. Theory, vol. 19, pp. 471-480, 1973.
[2] S. S. Pradhan and K. Ramchandran, “Distributed source coding using syndromes (DISCUS): Design and construction,” IEEE Trans. Inform. Theory, vol. 49, no. 3, pp. 626-643, Mar. 2003.
[3] D. Schonberg, K. Ramchandran, and S. S. Pradhan, “LDPC codes can approach the Slepian Wolf bound for general binary sources,” Proc. 40th Allerton Conf. on Commun., Control, and Computing, Monticello, IL, 2002.
[4] A. W. Eckford and W. Yu, "Density evolution for the simultaneous decoding of LDPC-based Slepian-Wolf source codes," in Proc. IEEE International Symposium on Information Theory (ISIT), Adelaide, Australia, pp. 1401-1405, 2005. [PDF]

Meeting of the minds

2011-05-17T23:53:00.000-04:00

At CWIT in Kelowna, here are the only three Canadians to be president of the IEEE Information Theory Society. From left to right: Professors Vijay Bhargava, Ian Blake, and Frank Kschischang.

Two papers at CWIT

2011-05-16T08:33:00.001-04:00

I have two papers at this week's Canadian Workshop on Information Theory in Kelowna:

1. Lu Cui and Andrew W. Eckford, "The delay selector channel: Definition and capacity bounds"
Session: Coding and Information Theory I, Wednesday May 18, 9:00-10:40 AM

This is work from Lu's master's thesis [PDF]. The "delay selector channel" is a discrete-time channel model that captures some of the features of molecular communication with Brownian motion. The main contribution of this paper is a closed-form lower bound on the channel capacity.

2. Josephine P. K. Chu, Andrew W. Eckford, and Raviraj S. Adve, "Distributed optimization of the Bhattacharyya parameter in wireless relay networks"
Session: Relay Assisted Communication, Thursday May 19, 2:00-3:20 PM

This is work from Josephine's Ph.D. thesis. We give an iterative, distributed solution to the non-convex multi-source, multi-relay resource allocation problem, where the objective is to optimize the Bhattacharyya parameter for each source's transmission.

Paper PDFs will be posted shortly. I will be in Kelowna and presenting both papers.

A quick exercise on divergent sequences

2011-05-09T14:09:00.002-04:00

I have a sequence and I'm trying to show that it converges. Here's my attempt to turn a morning of frustration into a blog post.

Let s(j), j = 1, 2, ..., be a sequence of real numbers with the following properties:

There exist constants a and b such that a <= s(j) <= b for all j.
In the limit as j goes to infinity, s(j) - s(j-1) = 0.

Conjecture: Any sequence s(j) satisfying these properties is convergent.

Your job: Disprove the conjecture by providing a counterexample.

There are many possible answers, but I give one in the comments.

Seminar in Ottawa: Information-theoretic problems in molecular and nanoscale communication

2011-04-24T13:51:00.001-04:00

I'll be in Ottawa this Thursday and Friday: Thursday I'll be the external examiner at the Ph.D. defense of one of Yongyi's students, and Friday I'll be giving a seminar on some of my recent molecular communication work. Details follow.

Time: Friday, April 29, 11 AM

Title: Information-theoretic problems in molecular and nanoscale communication

Abstract: Recent advances in MEMS/NEMS and systems biology have made it possible to manufacture customized nanoscale devices, such as swarms of nanorobots. However, manufacturing is not enough: a significant remaining challenge is to solve the communication problem among these devices, which would allow them to coordinate their actions. Furthermore, the nanoscale communication environment is rather different from the systems that are usually studied by information and communication theorists. In this talk, I will introduce the nanoscale communication problem from an information-theoretic perspective, focusing on molecular communication, which mimics the way in which microorganisms communicate. Mathematical models and achievable-rate results will be presented, and important open problems will be discussed.

A critic speaks

2011-04-19T09:32:00.001-04:00

Personally, I have never been able to convince myself that the Metropolitan Opera broadcasts are cultural in any deep sense of the word. The broadcasting of ludicrous Nineteenth Century melodrama, accompanied by music which is emotionally powerful but frequently of staggering vulgarity, and projected by means of acting which often falls below the level of a high school literary society's play, is cultural only on a very special level. I am devoted to opera and never miss a performance if I can help it, but if I were going to broadcast something that was cultural, I would certainly not choose Lucia, La Traviata or Tosca.

Yours sincerely,
Robertson Davies

(I was a resident at Massey College during my Ph.D., where Davies was the founding master. [Source, via])

Paper at MoNaCom

2011-04-11T15:21:00.003-04:00

I had a paper in the MoNaCom workshop, which was held on Sunday afternoon in Shanghai, in conjunction with Infocom 2011. Satoshi presented the paper; unfortunately I had to miss what looked like a very interesting workshop.

In our earlier work on microchannel molecular communication, we noticed that naive placements of the transmitter and receiver were suboptimal. However, when we tried to optimize these placements, we ran into a problem: our high-fidelity simulation of the molecular motors was computationally intensive, so it would have taken forever to run the huge number of simulations required in the optimization problem. Addressing this problem, our paper gives a quick-to-simulate, though approximate, model for motor trajectories, and applies it as a design tool for optimizing information rates in microchannels.

N. Farsad, A. W. Eckford, S. Hiyama, and Y. Moritani, "A simple mathematical model for information rate of active transport molecular communication," in Proc. 1st IEEE International Workshop on Molecular and Nano-Scale Communications, Shanghai, China, 2011. [PDF]

Random matrices and the Kronecker product

2011-04-08T12:27:00.007-04:00

The distribution of a zero-mean, jointly Gaussian column vector x is pretty basic stuff in probability: we get the covariance matrix R, given by

where the superscript T represents transposition. Then we find the probability density function (pdf)

where k is the number of elements in x.

But suppose you have not a random vector, but a jointly Gaussian, zero-mean k by k random matrix X. How do you express the pdf compactly? And can you compactly represent the pdf of matrix multiplications of X?

The easiest way to get the pdf of X is to turn X into a vector: for example, we can form the "matrix stack" of X, s(X), by stacking the columns of the matrix on top of each other as a column vector. That is, if X can be written

then s(X) is given by

Now, letting

we can compactly express the pdf of X as

Finally, we add the following twist: say we want to do some signal processing on the random matrix X. In particular, we want to calculate the matrix Y, given by

where A and B are u by k and k by v deterministic matrices, respectively (u and v can be arbitrary, but k must be the same as X's dimension so that the multiplication works out). Then Y is a u by v matrix.

To get the PDF of Y, we need to calculate the covariance matrix of its stack. All we know is the covariance matrix of s(X), so it would be most convenient to express it in terms of that.

To do this, we can use the Kronecker product, represented by ⊗. It is not too hard to show that the stack s(Y) of Y is given by

Thus, we have

and at last, f(Y) can be expressed in terms of s(Y) and s(Y)'s covariance matrix, just like f(X).

See also: A. Graham, Kronecker Products and Matrix Calculus with Applications, New York: Wiley, 1981.

Is my salary any of your business? (Updated)

2011-04-01T09:41:00.001-04:00

UPDATE: Doorey has an interesting take on the sunshine list.

For the first time, I made it onto Ontario's salary disclosure list. I won't link to it, but a trivial amount of Googling would lead you to find out exactly what I earned last year.

On the one hand, my employer is a public institution, and I agree that the public deserves to know how their money is being spent. And there are some individuals, like university presidents and senior executives of public corporations, whose compensation packages are large, and who negotiate their employment contracts individually; these packages should be on the public record. But I'm a middle-ranked professor with no administrative authority, and my pay is set by collective bargaining. Do you need to know exactly how much I, personally, get paid?

The disclosure list would perform its function if it was published as it is today, but with names redacted: the public would know, in general, how much professors and other civil servants get paid. I don't see what extra public good is served by printing my name, which is a significant invasion of privacy.

And a memo to the local papers: "Many more public servants earning six figures" is not news, although it certainly provides inflammatory fodder for people who think public servants are all fat cats helping themselves at the public trough. Salaries naturally go up as a result of cost-of-living increases, which track inflation, so it would be a huge surprise if the list did not grow. A much better question, whether the size of the list is growing at the expected rate given inflation, is not answered by any of the media reports I read.

How engineers look at the world?

2011-03-28T20:02:00.000-04:00

source: SMBC

The Tim Hortons School of Probability

2011-03-21T20:25:00.003-04:00

Do you know what this object is? We've just come to the end of that late-winter Canadian ritual when we line up to buy coffee, hoping for the magic words: "Win donut / Gagnez un beigne".

There are a couple of neat posts here and here, using "roll up the rim" to illustrate the binomial distribution. But you can also use it to illustrate the geometric distribution.

The geometric distribution (GD) gives a distribution on the number of trials before the next win. It's like the binomial distribution (BD) for impatient people: The BD asks, "If I buy 20 coffees, how many prizes can I get?" while the GD asks, "How many coffees do I have to buy until I get my next prize?"

More formally, let p represent the probability of receiving a winning cup, and let x represent the number of coffees you buy until you receive your next winning cup. Then since x follows the GD, we have

Rather than Pr(x), the probability of winning in exactly x cups, you're probably more interested in the probability of the next win happening in up to and including x cups. This is given by the cumulative distribution function F(x), where

Notice that Pr(x) forms a geometric series for x = 1, 2, ..., so using the geometric sum formula, we have

As we can see, the good news is that F(x) goes to 1 very quickly as x increases. This year's contest had published odds of p = 1/6, so using F(x), we have F(5) = 0.598; F(10) = 0.839; F(20) = 0.974; and F(30) = 0.996. That is, if you buy a coffee a day, your chances of winning within a month are over 99%.

The bad news, sort of, is this: say you bought a coffee today, and lost. From the GD, how long do you now have to wait before your next win? Measuring time x from before you bought the first cup, we now know that x > 1, since the first cup is a loser. Thus, we need to calculate the conditional probability Pr(x | x > 1). This is given by

That's very tidy. In general, you can show that Pr(x | x > m) = Pr(x - m) for any m > 0.

But now think of it this way: you already bought the losing cup, so how long do you have to wait after that until you win? Well, x is the time from before the first loss, so let y = x-1 be the time from after the first loss. Then from the above, Pr(y | x > 1) = Pr(y), the same as Pr(x). In other words, if you know you lost already, you still have the same waiting probabilities before your next win!

This seems perverse, but it actually makes sense: it's the memoryless property. Tim Horton (rest his soul) is not watching you buy coffee, and adjusting the order of cups as you win or lose. Instead, each play is independent and has the same odds. For the same reason, if you take a coin and happen to flip a million tails in a row, your odds of getting a tail on the next flip are still 1/2; the coin doesn't remember.

Just something else to think about next time you see the dreaded "Please Play Again / Reessayez SVP".

A quick update

2011-02-22T16:03:00.003-05:00

I'll be liveblogging the India trip over here. (Written from Pearson, waiting for my flight to Frankfurt.)

A very short presentation on email etiquette

2011-02-18T11:54:00.000-05:00

After receiving lots of questionable emails, today I gave a presentation on email etiquette to my first year class. I think they got the picture. Please feel free to use and/or modify it yourself.

PPT version, PDF version

Whirlwind Tour of India

2011-02-18T09:59:00.003-05:00

A couple of months ago, I agreed to join a delegation to India from York's Faculty of Science and Engineering. The idea is to make contact with Indian researchers, sign memoranda of understanding (MOUs) (e.g., covering graduate students and visiting faculty), and possibly to explore joint research projects through bilateral partnership programs.

We're going next week. My schedule includes a day of meetings at the Bose Institute and University of Calcutta on the 24th; then off to Mumbai for meetings at TIFR on the 25th and IIT Bombay on the 26th, then off to Delhi that evening; a day off on the 27th (the plan is to visit the Taj Mahal); and then a workshop in Delhi from the 28th to the 2nd, with side visits to IIT Delhi and JNU. It'll be my first trip to India.

The trip is partially funded by a government agency. On the one hand, this is great: research is increasingly an international activity, but research funding remains annoyingly nationalistic. In the absence of partnership programs like this one, it is almost impossible to construct international collaborative research teams, and breaking down those walls is a good thing. On the other hand, it's surprising how eagerly universities are jumping at this money: we've been told this is "delegation season" in Inda; we're arriving shortly after another Canadian delegation, and we're being told which other universities have signed MOUs, hoping not to miss out on our chance. You could argue this is symptomatic of general research underfunding in Canada; one wonders what Ghoussoub would have to say about it.

The blog is on hiatus for a bit, but I'll post about the trip when I get back.

How about a redesign?

2011-02-16T12:00:00.000-05:00

Everybody's doing it, so why not me? Hopefully this will go down better than theirs. I'll probably twiddle with the colors and so forth, but my main goal was to restrict the column width of the posts.

A famous communication theorist

2011-02-10T10:23:00.001-05:00

I was reading a paper on MIMO recently, when I noticed that one of the authors was Maher Arar. Surely it can't be that Maher Arar, I thought to myself. But in fact it is: here he is getting a PhD in EE from the University of Ottawa.

Congratulations, Dr. Arar! He even has a nice blog.

Pigeonholes and entropy, a work in progress

2011-02-03T20:43:00.001-05:00

The pigeonhole principle is probably the simplest lemma in combinatorics:

Say you want to sort k objects into n bins. If n < k, then at least one bin must contain at least two objects.

For example, a person can have only one of 366 possible birthdays (counting Feb 29). By the pigeonhole principle, any gathering of 367 or more people must have people who share a birthday.

The lemma is obvious almost to the point of triviality, so it comes as a surprise that you can use it to prove powerful results; there are some examples in the book.

Lately I've been wondering whether it can be used to say anything interesting about entropy. Here's the first thing I thought of: let X and Y be random variables on a discrete alphabet, and let S(X) and S(Y) represent the support of the probabilities of X and Y, respectively (i.e., x is in S(X) if and only if p(x) > 0). Then

Theorem. If |S(X)| > |S(Y)|, then H(X|Y) > 0.
Proof. By definition of entropy, H(X|Y) >= 0. If H(X|Y) = 0, then there must exist an injective map from S(X) --> S(Y). However, since |S(X)| > |S(Y)|, no such map exists (by the pigeonhole principle). Thus, H(X|Y) != 0, and the theorem follows.

Not very exciting, I will admit. I'm still thinking about it, any other ideas?

Bad web strategies: A short problem set

2011-01-27T09:07:00.003-05:00

Question 1. Consider this tweet:

What is the target of the link?

a) The ComSoc blog.

b) A link to a screenshot of the ComSoc home page, highlighting the "Blog" button in the link bar. The screenshot is part of a Facebook album, and clicking on the screenshot only navigates to the next picture in the album. Being an image, text in the screenshot (including the blog URL) is not selectable for copy/paste. From the link, there are no obvious ways to get to either the ComSoc home page or blog.

Question 2. What is ComSoc's slogan?

a) "The world's leading membership organization for communications professionals"

b) "We have never heard of the internet before today"

What's a PhD worth at the finish line? On hiring committees

2011-01-18T11:56:00.003-05:00

The blogosphere is roiling over the worth of a PhD. Should PhD programs take all qualified students? Is the academic "meritocracy" a myth, i.e., are some PhDs more equal than others? Even an outlet as erudite as The Economist suggests doing a PhD might be a waste of time (the always-nameless correspondent calling her ecology PhD "pointless").

In all of these pieces, the worth of a PhD is measured by its convertibility into a tenure-track position. There's lots of ink to be spilled on this topic, and I don't have a grand, philosophical point to add about the worth of the PhD. However, I'd like to talk about the "finish line" viewed by many PhD students: the hiring committee.

A few years ago, I had a chance to serve on a hiring committee. If you ever have a chance to do so as a student, I strongly, strongly recommend it -- it was extremely revealing to see what went on in the room, what the committee liked and didn't like, and the mistakes made by candidates, both on their applications and in their interviews.

Bearing in mind that this is a snapshot of one hiring process in one particular discipline, here's my memory of what the committee thought was important, in order from most to least important, in their decision making (I'm deliberately vague so as not to give away identifying details):

Publication record. Unsurprisingly, this was the sine qua non of the hiring process; the list of publications was our first stop in a candidate's CV. We were hiring in a particular sub-discipline, and we wanted to see a strong publication record in that area, both quantity and quality (i.e., top journals, first-name publications). Here, excess quantity without quality tended to count against an applicant, as did resume-padding (like listing a paper presentation at a conference with "invited talks", without explicitly making the distinction). We also had a few applicants in tangentially related areas, trying to convince us that their work was applicable to the sub-discipline in which we were hiring; these applications were quickly dismissed.
The interview. Our internal rankings changed significantly after the interview -- you might say that your publication record will get you on the short list, but the interview will get you the job. This one surprised me: I thought the interview would be a formality, but the best interviewees presented themselves as both good colleagues and strong communicators, whereas the worst did not. We were very surprised by some of our short-listers, both positively and negatively.
Reference letters. Important, but less important than you might think. Two reasons for this. First, the letters are predictable: you ask people who know you best for your letters, so most people had glowing letters from their PhD and post-doc supervisors. Second, hiring committees are broad whereas academic sub-disciplines are narrow, so -- except in a few extreme cases -- if you ask a superstar in your narrow field to write you a letter, few committee members would know the person. As I recall, letters would only hurt candidates, never help -- in one case, a letter was positive, but embarrassingly short; in another, someone got a superstar to write a letter (the one and only case of a superstar letter where I recognized the name), but it was clear that the person either didn't know or didn't think highly of the candidate.
Job talk. Surprisingly unimportant given the candidate's stress level. Before they walked through the door, we already had a very clear picture of the candidates' research programs. Further, we started the interview process at 9 AM, and the job talk happened in the afternoon, so by the time the job talk happened we already had an idea of the candidate's communication skills. And everyone always has a polished, well-prepared talk. So the job talk was about confirming impressions, rather than about any serious evaluation.
Where you got your PhD. Irrelevant, controlling for all the above. You got in to Harvard? Good for you. What have you done lately?

So is academic hiring meritocratic? I would say yes, for suitably constrained definitions of a meritocracy. Even in the sciences, a famous supervisor's name on a paper may ease its way into a top journal, and access to labs and high-quality equipment both depend on the university and the supervisor. But the most important facts about a candidate -- publication record and interview -- are generally about themselves, not about who they know or where they come from. In principle, there's no financial or personal barrier to publishing research in a top journal; one need only have a good idea. And seeing all the applications together, it was pretty obvious which researchers were talented and which were not; there was discussion of course, but ultimately the rank ordering of candidates was not terribly controversial among our committee.

No, I don't want to "talk about" your exam

2011-01-10T14:21:00.003-05:00

The college where I did my undergrad had the habit of printing your rank in your class along with your yearly grade report; the top student in each class got a small prize, usually a book. At the end of my third year, I was happy to read on my report that I ranked first out of thirty-something electrical engineers (it was a small college).

I left for the summer and returned for my senior year, but hadn't received my prize. When I asked around, it turned out that the second-place finisher had complained to all our profs about his exam grades, and managed to get enough extra marks to take first place -- my book became his.

I realize it's a small, petty thing to remember after all these years. But I think about it every time a student comes to my office wanting to "talk about" their mark on the final exam.

A look back

2010-12-18T10:12:00.000-05:00

At the beginning of the year, I resolved to blog more seriously in 2010. I didn't quite hit my goal of 52 blog posts, but I'm still pretty happy with how it turned out.

This blog will be on hiatus until the new year. Happy holidays, and see you in 2011.

Two new papers on molecular communication

2010-12-02T20:50:00.003-05:00

I just got back from lovely Cambridge, Mass., where I managed to sneak in a day to see some of the interesting talks at Bionetics before storming back to Toronto to cover my teaching schedule. December in Boston, what more could you ask for?

I'm a coauthor on a couple of new molecular communication papers that just came out, one at Bionetics, and the other on arXiv:

For Bionetics, my Ph.D. student, Nariman Farsad, and our collaborators, Satoshi Hiyama and Yuki Moritani of NTT DOCOMO, continued some of our earlier work on microchannels. This was a short "work-in-progress" paper, where we considered the performance of active transport systems in a more realistic simulation environment. Surprisingly, Brownian motion with drift usually does better than molecular motors -- probably because every molecule can start propagating right away, rather than waiting to be picked up by a motor. [N. Farsad, A. W. Eckford, S. Hiyama, and Y. Moritani, "Information rates of active propagation in microchannel molecular communication,” in Proc. 5th International Conference on Bio-Inspired Models of Network, Information, and Computing Systems, Boston, MA, USA, 2010. (pdf)]

On arXiv (and also submitted to Trans. Info. Theory), K. V. Srinivas, Ravi Adve, and I extended Sachin's earlier work on Brownian motion with drift. We're looking at molecular timing channels, where information is encoded in the transmission time x. The receiver sees the reception time y, where y = x + n, and where n is the first arrival time of a Brownian motion; thus, these are additive noise channels. It turns out that the first arrival time distribution of a Brownian motion with positive drift is given by the inverse Gaussian distribution (kind of an odd name at first glance, since it's not the reciprocal of a Gaussian PDF). So a molecular timing channel in the presence of positive drift is an additive inverse Gaussian noise channel. There is a rich literature on the inverse Gaussian, which lets us get closed-form solutions and bounds for a number of important quantities, like ML detectors, probabilities of error, and capacities -- sadly, not as clean and nice as the solutions for the AWGN, but we do what we can. [K. V. Srinivas, R. S. Adve, and A. W. Eckford, "Molecular communication in fluid media: The additive inverse Gaussian noise channel," arXiv:1012.0081 (submitted to IEEE Transactions on Information Theory).]

You can also check out more of my molecular communication work here.

Sixty years of Hamming codes

2010-11-25T20:54:00.000-05:00

Via Slashdot, this year marks the 60th anniversary of the venerable Hamming code. Arriving just two years after Shannon's seminal paper, the Hamming codes were the first nontrivial error-correcting code, and Hamming's contribution to the theory of error correcting codes (using metrics which are now known as Hamming distance and Hamming weight) are still taught today.

Here (pdf) is Hamming's original paper (from BSTJ), and here is my Hamming code lecture from last spring's information theory course:

Ph.D. position available (Update: No longer available)

2010-11-12T11:51:00.003-05:00

Update August 16, 2011: This particular position is no longer available. However, we may have openings in related positions. Please see here before you apply. Original post follows.

My lab will have a Ph.D. position available in September 2011 in the area of biomechanical signal processing. I've been doing some work with body-area networks composed of accelerometers, and the Ph.D. position will be to continue that work in a more theoretical direction, with mathematical modeling, simulation, and analysis. The position is interdisciplinary, and you would be co-supervised by Will Gage in Kinesiology.

The official "help wanted" ad is below (please do me a favor and share it widely). If you're interested in applying, please contact me first. (So I don't get you confused with the reams of generic Ph.D. applicants that I get, please put "Biomechanical Ph.D. position" in your subject line.)

(PDF version)

Capacity in the presence of intractability

2010-11-10T10:31:00.001-05:00

Shannon's channel capacity, C = max_p(x) I(X;Y), gives the highest rate at which information can be reliably transmitted over a noisy channel. But the simplicity of the equation is deceptive: there are very few channels for which capacity is known. Normally, the difficulty is in optimizing over all input distributions p(x), but what if I(X;Y) is itself intractable?

I(X;Y) can be hard to calculate if f(y) or f(y|x) are hard. The former can happen in intersymbol interference channels with long memories, especially under a binary (or discrete) input constraint. The latter can happen in timing channels, when x is a vector of inputs, and the output is of the form y = sort(x+n), where n is the vector of timing noise. In this case, calculating f(y|x) requires the calculation of a matrix permanent, a classic intractable problem for vectors of practical length. (See also: the Bapat-Beg theorem.)

However, there is an elegant trick you can use to find a lower bound on I(X;Y), as long as you can (tractably) generate random variables x and y according to their true distributions f(x) and f(y|x). Let g(y|x) be any valid conditional distribution of y given x. Further, let

Then

where expectation is always taken with respect to the true distribution. That is, if you can pick a tractable approximate distribution g(y|x), and you can simulate both x and y, then you can perform a Monte Carlo expectation on log g(y|x)/g(y), and get a lower bound on I(X;Y). You can then optimize with respect to the bound to find a lower bound on capacity.

The proof is straightforward. First let g(x|y) = g(y|x)f(x)/g(y). Then:

So the accuracy of the bound is equal to the relative entropy between the true f(x|y) and the approximate g(x|y).

This result has been around for a while, but a nice recent application is here:

D. M. Arnold, H.-A. Loeliger, P. O. Vontobel, A. Kavcic, and W. Zeng, “Simulation-based computation of information rates for channels with memory,” IEEE Trans. Inform. Theory, vol. 52, pp. 3498–3508, Aug. 2006.

List of blogs in machine learning

2010-11-02T15:53:00.000-04:00

Here is a thread with a huge list of good machine learning blogs.

The most interesting discovery from the list: Radford Neal, the co-re-discoverer of LDPC codes, has a blog. I took his graphical inference course when I was a Ph.D. student at U of T.

Announcing the Nano Networks blog

2010-10-26T08:52:00.004-04:00

As one of my volunteer service positions, I am chair of the IEEE ComSoc Emerging Technologies Subcommittee on Nanoscale, Molecular, and Quantum Networking. I've decided to transform our committee's home page into a blog, to share research news and other details about this field.

You can read the blog at http://nano-networks.blogspot.com/. You can also subscribe to our Twitter feed: @NanoNetworks.

The blog will be newsletter-style; if you have anything interesting to post (new paper, profile of yourself or your lab, upcoming conference, or anything else), please send me a short blurb, preferably something ready to post. I will try to make use of anything sent to me.

Entropy fact of the day

2010-10-14T09:53:00.007-04:00

It's well known that certain distributions have maximum entropy under moment constraints. For instance, maximum entropy H(X) under a variance constraint is given by the Gaussian distribution. Generally, for any positive integer k and for given constant c, there exist algorithms to find the distribution maximizing H(X) subject to E[X^k] = c.

However, here's what I learned yesterday: if we change the constraint to E[X^-k] = c, there exists no such maximum entropy distribution -- because you can always achieve unbounded entropy subject to the constraint.

At first I thought this was surprising, but then I realized it's not too hard to think of an example that shows why it's true. Think about it for a few minutes -- I'll post a hint in the comments.

How to get my attention

2010-10-01T09:49:00.002-04:00

Every couple of months or so, a wave of emails from prospective PhD applicants will flood my inbox. A typical email will go something like this:

Dear Professor Andrew Eckford,

I saw your web page and am very interested in your research on computer networks. I have the same research interests as you and would like to apply for a PhD position under your supervision. My master's thesis was entitled "[topic I'm not interested in]" and I published my results in [conference/journal I've never heard of]. I am attaching my resume to this email and look forward to your positive reply.

The messages are so similar, both in terms of content and geographic origin, that I have to think there is some agency behind them. But particularly because they are so similar, it's easy to dismiss them as spam and ignore them. It probably doesn't hurt an applicant to send these emails, but I would never remember such a message at admission time, so it certainly doesn't help either.

What would get my attention is an applicant who makes it clear that s/he is not just interested in the PhD slot. In fact it's probably unnecessary to mention in the first email that you're a Ph.D. applicant; what I really want are bright students who are interested in research in general, and my research in particular. I've never received an email like this from an applicant, but I would certainly remember it if I did:

Dear Professor, I read your recent papers on fractional cooperation, and thought the idea was interesting. I have the following comments and questions about your scheme ... [some insightful commentary follows].

Of course, a more personal touch works for me, because my group is small and I can afford to be very selective with my Ph.D. applicants. Maybe the form letter approach works for profs whose groups are measured in the dozens.

Embedding fonts on Mac (or, another reason why I love my MacBook)

2010-09-18T20:06:00.004-04:00

I just learned something remarkable, and I have to write it down so I don't forget:

Often IEEE conferences will require you to submit a PDF with embedded fonts. If you're forced to do this, no doubt you're familiar with a hugely irritating fact about most LaTeX packages: they don't embed fonts by default. There are workarounds, like the ps2pdf command line parameters that force font embedding, but using them can be awkward (e.g., if your LaTeX target isn't postscript, or if you're using a GUI LaTeX editor).

But if you're a Mac user, like me, here's all you have to do: Open the PDF in Preview, and then save it again. It's saved with fonts embedded. That's it, seriously!

Odds and ends

2010-09-17T09:01:00.003-04:00

Odds and ends from the past week:

If you're the kind of person who turns off WiFi SSID broadcast because you think it makes your network more secure, don't bother.

Once in a while, I write an exam question which compares the data rate of an internet connection with, e.g., physically carrying a hard drive from place to place. Turns out somebody did this in real life using carrier pigeons. Result: in rural England, pigeons with microSD cards beat the internet by a wide margin. (via)

Good news: Edward Burtynsky was hard at work this summer in the Gulf of Mexico. (via)

PhD positions open at UIC

2010-09-14T09:54:00.005-04:00

I got the following last night. For any senior EE undergrads or master's students out there, this is a great opportunity to do some interesting information theory work. (Let me also say that Chicago is one of my favorite cities in the USA.)

Click the poster to see full size, or click here for the PDF version. I don't have any more details than are in the announcement -- contact the professors directly for more info.

A question about physical layer, quote, security, endquote

2010-09-09T20:33:00.010-04:00

Physical layer security sure is hot these days. Its proponents claim provable security, something the cryptographic community hasn't yet been able to provide. Sounds great!

Physical layer security is largely based on the wire-tap channel; here (pdf) is one of the seminal papers on the subject. The great achievement in the wire-tap channel is to allow transmission at capacity on the from source to the legitimate receiver, while the mutual information from source to wire-tapper is zero, even if the wire-tapper knows the code book. Thus, we have "proven" that communication is secure, because the wire-tapper can never accurately guess the message that was sent to the legitimate receiver.

But here's the thing. Look at figures 1 and 2 in the paper I link. Everything relies on the wire-tapper having a physically degraded channel with respect to the legitimate receiver. This makes sense: if, somehow, the situation were reversed and the legitimate receiver were degraded with respect to the wire-tapper, it would obviously be impossible to prevent the wire-tapper from decoding the messages. Put another way, there is no security unless the wire-tapper has a worse channel than the legitimate receiver.

Here is my question: isn't it misleading to call this "secure"? It is unlikely that the wire-tapper would oblige us by providing his channel state information. Thus, we merely exchange one set of uncertainties for another: namely, exchanging uncertainty about the hardness of the factoring problem for uncertainty about the wire-tapper's channel state -- except that factoring is very widely believed to be intractable, whereas it's not hard to imagine a committed adversary being able to find a good channel for a wire-tap.

One up, one down

2010-09-04T19:49:00.003-04:00

Maxim Raginsky kicks off his new blog, The Information Structuralist.

Meanwhile, Mitzenmacher signs off his blog to focus on his new administrative appointment.

(Raginsky's blog via)

Five-Minute Info Theory Problem

2010-09-02T09:43:00.004-04:00

I can't believe it's September already. Astonishingly, I managed to get eight of my nine summer to-do list items done, at least partly.

Here's a simple problem that came up yesterday when I was looking for a quick bound on entropy ... I thought it would be a nice warmup to kick off the term:

Let A and B be independent random variables. WLOG, suppose H(A) <= H(B).

Show that H(A) <= H(B) <= H(A+B) <= H(A)+H(B).

Post your answer in the comments.

Hiatus

2010-08-19T13:26:00.001-04:00

No posts for a while. See you in September.

Paper at IEEE NANO: Brownian motion versus motors for molecular communication

2010-08-17T10:15:00.007-04:00

IEEE NANO is happening this week, and I have a paper:

A. W. Eckford, N. Farsad, S. Hiyama, and Y. Moritani, “Microchannel molecular communication with nanoscale carriers: Brownian motion versus active transport,” in Proc. IEEE International Conference on Nanotechnology, Seoul, Korea, 2010. [PDF] [Poster PDF]

Building on my earlier molecular communication work, this paper looks at communication in microchannels, like you would find in a lab-on-chip device. In these devices, molecules can propagate either randomly by Brownian motion, or by piggy-backing on a molecular motor. We compared the information rates achievable by each, and found two distinct regimes where each method works best. When the number of information-bearing molecules is small, motors are superior, since their motion is much less random than Brownian motion. However, for large numbers of molecules, Brownian motion is best, because all the molecules can start propagating immediately, rather than waiting for a motor to pick them up.

This was joint work with my Ph.D. student, Nariman Farsad, as well as Satoshi Hiyama and Yuki Moritani from NTT DOCOMO. Nariman is presenting the paper.

Making the list

2010-08-13T11:41:00.005-04:00

In case you're wondering, the 2010 Academic Ranking of World Universities (PDF) is out. How'd we do? York is in the 401-500 category alongside Canadian universities like Carleton, Concordia, Sherbrooke, and the Université du Québec system. Some other schools in the category were Boston College, Lehigh University, and the University of Tehran.

Altogether there were 23 Canadian universities in the top 500, with U of T topping the Canadian contingent at 27th spot.

Of course there are lots of problems with university rankings -- probably more meaningful to compare individual departments, and even then not so much. But it's nice to be on the list anyway.

(List via @InklessPW.)

P vs. NP: Solved? (Update: No)

2010-08-09T12:37:00.020-04:00

There's a paper under review claiming to prove that P ≠ NP. The claimed proof is being treated credibly: Stephen Cook is quoted as saying, "This appears to be a relatively serious claim to have solved P vs NP." Dick Lipton, who maintains a blog on complexity theory and who has a summary of the paper, is optimistic but points to some possible problems with the proof.

Complexity touches information theory at a number of points. For instance, many optimal detection and error-control coding problems are known to be NP-hard: syndrome decoding of a linear code (1), minimization of the trellis complexity of a linear code (2), and optimal multiuser detection (3) are good examples.

Update (Aug. 11): This BBC article doesn't add anything new, but has one of the nicest layman's descriptions of P versus NP that I've ever read:

Scott Aaronson, a computer scientist at the Massachusetts Institute of Technology (MIT) in the US, explained to BBC News why this problem was so significant. "People sometimes use the analogy of a jigsaw - it can be hard to complete the jigsaw, but if someone has done it, it's pretty easy to check - you just look at it," he said.

P vs NP poses the following question: If there is a problem that has this property - whereby you could recognise the correct answer when someone gives it to you - then is there some way to automatically find that correct answer? "There's always one way a computer can find the answer- just by trying all the possible combinations one by one," said Dr Aaronson. "But if you're trying to break a cryptographic code, for example, that could take an astronomical amount of time. P vs NP is asking - can creativity be automated?"

Meanwhile, the theoretical CS community is going bonkers over this result. Aaronson has bet $200,000 that the proof will not stand up to scrutiny. Lipton has collected the four main objections in early readings of the proof so far, and there is now a wiki devoted to analysis and criticism of the proof.

Also, from here, a nice quote about proofs in general: "A proof only becomes a proof after the social act of 'accepting it as a proof'." -- Yuri Manin

Update (Aug. 12): In a comment on Lipton's blog, the awesome Terry Tao more or less declares the proof to be toast. Aaronson regrets his bet, even though his money is most likely safe.

Final update (Aug. 13): A highly technical post on Lipton's blog (which I don't pretend to understand at all) seems to torpedo the proof. The post's commenters agree.

On rejection

2010-07-28T09:03:00.003-04:00

Early on, a major issue with evolutionary theory was the lack of "early" animal fossils: complex animals, like trilobites, were well known in the Cambrian fossil record, but there was no evidence of their evolutionary precursors.

All this changed in 1946, when geologist Reginald Sprigg made a startling discovery. The New York Times has the story:

Reginald Sprigg, a geologist for the South Australia government, was checking out some old mines in the Ediacaran Hills of the Flinders Range several hundred miles north of Adelaide. Sprigg noticed some striking disc-shaped impressions up to four inches in diameter on the exposed surfaces of rocks nearby. Sprigg interpreted the patterns as the [pre-Cambrian] fossil remains of soft-bodied creatures like jellyfish or their relatives.

The kind of career-making discovery a young scientist could only dream of! But the story goes on (emphasis added):

When Sprigg first showed the imprints to leading authorities, they dismissed them as artifacts made by the weathering of the rocks. ... Later that year, when Sprigg found the frond-like forms he called Dickinsonia, he was certain that such geometrical impressions could have been made only by living creatures. But despite their potential importance, Sprigg’s discoveries were ignored at an international geology meeting and his paper describing the fossils was rejected by the leading journal [Nature]. Sprigg moved on to other, more rewarding pursuits in the oil, gas, and mining industries.

It took another decade for Sprigg's earth-shattering contribution to be widely recognized. Something to keep in mind both as an author and as a reviewer.

The other Blackwell channel

2010-07-23T09:30:00.010-04:00

In memory of the late David Blackwell (see obits here and here), I'd like to talk about the Blackwell channel. Not this one, which is the one everyone thinks of, but the other one -- otherwise known as the trapdoor channel, the billiard-ball channel, or the chemical channel. (Find me another information theorist with two channels named after them.)

The channel works like this. Information is transmitted using coloured balls (red to represent 0 and black to represent 1, say). The "channel" is a bag containing a single (red or black) ball. At each channel use, the transmitter inserts a single ball into the bag, and the receiver then removes one of the two balls, selecting uniformly at random between the two. It should be obvious that this model can be generalized in various ways.

Blackwell proposed this model in his 1961 book as an example of a channel with memory: the channel output is clearly dependent on prior channel inputs. Remarkably for such a simple model, capacity is still unknown. As far as I know, the best result is the channel capacity with feedback, found by Permuter et al. to be the logarithm of the golden ratio. Since the transmitter is free to disregard the feedback, this must also be an upper bound on the capacity without feedback.

This channel also admits zero-error codes. For example, even if the initial channel state is unknown, a threefold repetition code (i.e., red, red, red or black, black, black) can be decoded without errors, since there can be at most one wrong-coloured ball at the output; thus, the zero-error capacity is (trivially) at least 1/3. Zero-error capacity is also an open problem for this channel.

Why the rage against Engage?

2010-07-12T10:54:00.006-04:00

The Engage grant is a new NSERC initiative for kick-starting collaborations between industry and academia. The main features of these grants are: short duration (6 months max), limited value ($25,000 max), industry focus (an industrial partner, who will own all the IP, is mandatory), and quick decisions (6 week turnaround, which is only possible with no peer review). I found out on Friday that my Engage proposal would be funded; I'll write more about it as we publish, but the project is in the broad area of fractional resource sharing for femtocells, similar to the earlier work we've done on fractional cooperation (e.g., 1, 2, 3).

So imagine my surprise to learn that the CAUT has come out strongly opposed to Engage, passing the following resolution:

NSERC must remain a granting council that enables peer-reviewed fundamental research. Whereas linkages between industry and universities consistent with academic freedom are possible, the targeting of granting council funds to private industry’s needs erodes Canada’s capacity to contribute to the general advancement of knowledge in the public interest.

Of course I'm biased because I'm a recipient of NSERC's largesse under this program. But CAUT's position is bizarre on a number of counts. First, it treats all scientific research as equally separate from industrial applications. In wireless communications, there is no tension between the "general advancement of knowledge" and "industry's needs" -- they are one and the same, because essentially all of the applications are industrial. Second, it assumes that only non-industrial research is in the public interest. Again, in wireless, it's the opposite -- the type of research that has a real impact, say by improving your mobile's performance (like this), is coming out of industrial labs, not out of academia. Third, CAUT has completely missed the point of this program: its limited scope, duration, and funding makes it only useful as a door-opener, to give a researcher an easy way to approach a company, and to give that company a low-risk way to say yes to a collaborative project. What's more, Engage grants can only be approved for companies and researchers that have never collaborated before, so this granting program can't be part of some great outsourcing of corporate R&D to universities.

My view on these projects is broadly held across engineering faculties. So CAUT is misunderstanding its membership by opposing the Engage grant, and promoting a "pure" vision of academia that is potentially harmful in some disciplines.

Molecular communication using Brownian motion with drift

2010-07-06T15:31:00.005-04:00

I'm third author on this paper, and Sachin and Ravi rightly deserve most of the credit. However, we got a writeup in Technology Review, so I thought I would blog a bit about the paper.

In molecular communication, a transmitter sends a message to a receiver by releasing a pattern of molecules into a shared medium. One way to communicate using molecules is to use timing -- i.e., releasing a molecule at different times to express different messages. Using Brownian motion, the timing message is distorted by the random propagation time from transmitter to receiver.

My original work on molecular communication (1, 2) considered the case of drift-free Brownian motion. In this paper, we extend these results into Brownian motion with drift -- it's a much easier case to deal with, since the first arrival distribution without drift has a very heavy tail, as well as infinite mean. We also have some nice results about modulation: even when multiple molecules are available, pulse position modulation turns out to work quite well if the drift velocity is low.

The paper was submitted to IEEE Trans. Nanobioscience, and we're expecting the first reviews back any day now.

Just like being there

2010-06-22T09:30:00.000-04:00

I didn't catch ISIT this year -- my first miss since 2003. Sarwate (1, 2, 3, 4, 5) and Mitzenmacher (1, 2) have some nice summaries. Congrats to Shlomo Shamai for his Shannon Award win.

You should be tweeting

2010-06-03T11:55:00.001-04:00

Surprisingly few academics in my field are on Twitter. I'm one of them, and did my share of tweeting through ICC. Other than me, only four other people used the #icc2010 hashtag: @jabriffa, @csgrad, @gvrooyen, and @mayanm. This is at a conference with around a thousand attendees! Why the low uptake? It's not as though Twitter is brand new.

At ICC, it struck me that Twitter would be a great tool for participants to get the most out of large conferences. With 20 parallel sessions and over 1000 accepted papers, ICC is enormous -- it would be physically impossible to see more than a small fraction of the available presentations. To deal with this overload of information, Twitter would be an ideal tool to quickly find out about interesting sessions. For example, say you're at a huge conference: tweeting audience members could be giving real-time information on which sessions are worth attending, which presentations are the most engaging, and which papers might interest you. You might change your plans to swing by and check out a session that seems interesting -- or if you didn't have time, you could make a note and check out the papers in the proceedings. Otherwise, seeing these tweets would be a nice way to find out about researchers with interests similar to yours -- and for them to find out about you.

This networking effect of Twitter is a key feature that goes beyond conferences, especially for junior faculty and graduate students. Research -- even excellent research -- doesn't usually attract attention to itself; especially in a huge field like communications, it's incredibly hard for individual researchers to boost their signal above the background. As a result, researchers need to be advertising their work all the time, using every tool they can find. Twitter is ideal for this -- for instance, whenever I publish a paper or give a talk, I mention it on my Twitter feed, and it is broadcast to all my followers. (Of course, the trick is then to get lots of followers.) This principle applies to all social media, not just Twitter: I maintain a personal website, this blog and a YouTube page as well as Twitter, to make it as easy as possible for people to find out about my work.

As one commenter on Mitzenmacher's blog put it, any graduate student wanting an academic career should have a blog. I'd go further: graduate students and junior faculty should be blogging, tweeting, and doing whatever else they can to get their name and work out there.

Paper at ICC: PDF and video

2010-05-31T13:35:00.005-04:00

I'm back from beautiful Cape Town, where I presented the paper I mentioned earlier.

Citation:

N. Farsad and A. W. Eckford, “Resource allocation via linear programming for multi-source, multi-relay wireless networks,” in Proc. IEEE International Conference on Communications (ICC), Cape Town, South Africa, 2010.

PDF of the paper is here.

Presentation video (this is a playlist):

Seven continents

2010-05-25T01:41:00.015-04:00

It's long been a life goal of mine to visit all seven continents, which I managed to cross off the list when I stepped off the plane in Africa.

Pictures from: Australia (Sydney, Australia, 2005); Europe (Fontainebleau, France, 2008); Asia (Great Wall near Beijing, China, 2008); Africa (Cape of Good Hope, South Africa, 2010); South America (Colonia, Uruguay, 2006); North America (Banff, Canada, 2008); Antarctica (Weddell Sea, 2006).

Paper at ICC

2010-05-17T10:18:00.004-04:00

UPDATE: Paper PDF and presentation video are here.

I'll be heading to Cape Town later this week to present a paper at ICC:

Resource Allocation via Linear Programming for Multi-Source, Multi-Relay Wireless Networks
N. Farsad and A. W. Eckford
CT05: Tuesday, May 25, 1400-1545, Roof Terrace A

This paper continues my work on fractional cooperation, in which relay nodes only retransmit a fraction of the source's packet. Since the transmission is corrupted both by the channel effects and the possibility that not every source bit is retransmitted, we use LDPC codes to recover the original transmission. There are two key insights here: first, in LDPC decoding, the log-likeihood ratios of the relays are additive, weighted by coefficients equal to the source fractions (i.e., a linear combination); and second, for sufficiently many relaying partners, the sum of the LLRs has approximately a Gaussian distribution. Similarly to the design problem solved by EXIT charts, the problem of assigning source fractions can be posed as a linear program.

I'll be presenting the paper. Video of the talk and the PDF of the paper will be posted after the conference.

Paper at Queen's: PDF

2010-05-14T10:00:00.006-04:00

It was a beautiful day in Kingston on Wednesday, which I mostly spent indoors listening to talks.

Here are the PDF and citation of the QBSC paper I mentioned earlier:

S. E. T. Hadley, A. W. Eckford, and W. H. Gage, “Power saving in a biomechanical sensor network using activity detection,” in Proc. 25th Biennial Symposium on Communications, Kingston, ON, pp. 173-176, 2010.

Summertime, and the living is busy

2010-05-10T10:57:00.005-04:00

I love it when I tell someone that I'm a professor, and they respond with something like, "Oh, it must be great to have the whole summer off!" Especially today: with my spring grading finally out of the way, I'm making up my lengthy summer to-do list. Some highlights:

Clean up my office disaster
Attend this conference (and attend an executive committee meeting)
Attend and speak at this conference
Finally get the revisions done for this paper (which should have been done months ago)
Write a journal version of this paper
Get a grant proposal done
Read and edit theses as two students get ready to defend
Get some chapters done for a textbook project that were promised for months ago
Make some headway on a new consulting contract
Lots more that I'm forgetting, for sure

Yep. Another lazy, coronas-on-the-beach summer.

Paper at Queen's

2010-05-04T08:58:00.003-04:00

My paper at the Queen's Biennial Symposium on Communications:

Power Saving in a Biomechanical Sensor Network Using Activity Detection
Scott E. T. Hadley, Andrew W. Eckford, and William H. Gage
Wednesday, May 12, 4:50-6:10 PM (second talk in session), Walter Light Hall rm. 210

This is the first paper out of an interdisciplinary project with Will Gage in Kinesiology. The basic idea is to place accelerometers on stroke patients to monitor their recovery in walking, but the main challenge is to design a system with the robustness, endurance, and ease of use to be used in the home, rather than in the therapist's office. It's a practical project with lots of neat signal processing, and we just got a grant to continue it.

One of the key issues is to reduce power consumption and extend battery life. In this paper, we show that signal processing can help with power saving: we use activity detection to scale the sampling rate based on what the patient is doing. As a result, we strike a nice balance between power saving and data loss: switching modes, we lose about a second of data (about one step -- because it takes that long to detect acceleration with the reduced sampling rate), whereas we would lose about 30 seconds by putting the node to sleep and waking it up.

Scott will be presenting the paper. I'll be at Queen's for Wednesday only, so I hope to see you there. PDF of the paper will be posted shortly.

What's the worst noise?

2010-04-19T13:12:00.006-04:00

It's a well-known theorem of information theory that, of all additive noise channels with constant noise variance (i.e., constant noise power), the Gaussian channel has the minimum capacity. I learned this long ago, and over time the concept was distilled in my memory down to "Gaussian is the worst noise".

I was trying to use this "property" to find a quick lower bound for the capacity of a new binary-input channel. But after some funny results, my grad student found an old paper of Shamai and Verdú, which -- much to my surprise and dismay -- shows that Gaussian noise is not worst-case if the input distribution is constrained! In particular, if the input is binary {+1,-1}, the worst constant-power distribution is supported on the discrete set {...,-2,0,+2,...}, and can't generally be expressed in closed form.

For more details, see:

S. Shamai and S. Verdú, "Worst-case power-constrained noise for binary-input channels," IEEE Trans. Info. Theory, vol. 38, no. 5, pp. 1494--1511, 1992.

Video: Writing on Tiny Paper

2010-04-07T13:25:00.001-04:00

Here's the video for my talk at UIUC on Monday (this is a playlist). More details here.

Steve Munro Misses the (Swan?) Boat

2010-03-31T13:26:00.005-04:00

Some Toronto-specific stuff now. There are days when I wonder whether Steve Munro is this city's most overrated blogger, today for example: here he is arguing that the Toronto Board of Trade was full of it by saying Toronto’s commute times are the worst in the world.

Go read Munro's post and see if you can piece together his rambling argument. The best I could do: the Board of Trade study is wrong because it used a metric, namely average commute time, which under-emphasized proper urban development. But development is important, and development goes hand in hand with transit, so the government should invest in transit.

It's important to bear in mind that the report was hugely in favor of transit investment, stating unequivocally:

... congestion threatens Toronto’s viability over the long-term and serves as an argument for increased investment in public transit and policies that encourage Torontonians to leave their cars behind.

So Munro is spectacularly missing the point on two fronts. The first is the current political context: the provincial government just postponed (and maybe killed) Transit City, of which Munro is a staunch promoter. The Board of Trade report, coming on the heels of this announcement, is an obvious political win for Transit City advocates. Other commentators are picking up on the obvious ways to shame the province, and keep the issue in the public eye. Not Munro, though - he’d rather quibble about metrics, and take a far more convoluted route to argue about transit funding.

The second is that Munro completely misunderstands the report, by failing to see the value of time. The economic impact of commuting is most acutely felt in the time spent commuting, not in the distance or speed, since a long commute is both time and effort diverted from other (potentially money-making) tasks. If you’re going to argue that the province should invest in transit, surely the return on the investment should be largely economic, especially since an economic boost will return to the province’s coffers in the form of tax revenues.

Talk at UIUC: "Writing on Tiny Paper"

2010-03-29T12:39:00.002-04:00

I'll be giving a seminar next Monday, April 5, at UIUC. Thanks to Todd for the invite. Here are the title and abstract:

Writing on tiny paper: Towards information-theoretic analysis of molecular communication

In molecular communication, messages are transmitted as a pattern of molecules propagating from a transmitter to a receiver. This form of communication is useful where electromagnetic communication is impractical, such as in a lab-on-chip device. In this talk, we review current approaches to molecular communication. We also present some of our own results on modelling and analyzing these channels from an information-theoretic perspective, looking at molecular channels both as timing channels and as mass-transport channels.

Interview on TalentEgg

2010-03-26T13:49:00.002-04:00

At the beginning of the term, TalentEgg contacted me for an interview on my use of social media in the classroom. They posted the interview to their YouTube channel a few weeks ago, but I just noticed it today:

(I wish she told me to comb my hair ...)

Tenure

2010-03-12T09:18:00.002-05:00

Yesterday, I was informed that my application for tenure has been approved by the university president. I'll be an Associate Professor effective July 1.

Once again, Futurama shows us how to enjoy tenure properly:

Mayor: "Dr. Wernstrom, can you save my city?"
Wernstrom: "Of course, but it'll cost you. First I'll need tenure."
Mayor: "Done."
Wernstrom: "And a big research grant."
Mayor: "You got it!"
Wernstrom: "Also, access to a lab and five graduate students...at least three of them Chinese."
Mayor: "Did...all right, done. What's your plan?"
Wernstrom: "What plan? I'm set for life. Au revoir, suckers!"
Leela: "That rat! Do something!"
Mayor: "I wish I could, but he's got tenure."

Student rates

2010-03-08T19:00:00.004-05:00

Recently, Mitzenmacher had a nice post about conference budgeting. One interesting tidbit: this year's registration fee for STOC will be around US$500, and he asks, "At what point do registration fees become a noticeable concern?" Many commenters agreed that $500 was outrageous; meanwhile, in the IT community, last year's ISIT registration would have set you back a cool US$850.

However, concerning student registration rates, Mitzenmacher also asks the following question: "... every student who attends is actually a loss, that has to be covered from elsewhere. Is this the right way to go?"

At first, it seems like the answer should be: yes, of course; helping students is good. But thinking back to my graduate student days, I never paid for conference registration out of my own pocket -- it was always covered by my supervisor. Thus, the primary beneficiary of a student rate is a professor with many graduate students attending the conference -- who, as a result of his/her large group, is already well funded!

In fact, if student rates are offered at a loss, that loss is made up by professors (like me) with limited funding, who can only afford to send themselves and maybe their one student. So are student rates actually regressive?

You learn something new every day

2010-03-01T22:31:00.004-05:00

Did you know that (x!)^(1/x) approaches x/e + (const) for sufficiently large x? Wolfram Alpha says so.

Now I'm looking for a proof. Suggestions would be welcome.

UPDATE: Wrong! Alpha was right, I misinterpreted the result. Looks more like x/e + O(log x).

Why is ITA invitation-only?

2010-02-25T22:53:00.005-05:00

I was reminded today that ITA came and went earlier this month, so I've kept alive my five-year-long streak of not attending.

By all accounts, ITA is now a major event on the information theory calendar: the participants page lists over 350 attendees -- maybe a bit less than half of an average ISIT. And the ITSoc boldface names are attending; insofar as the meeting-and-greeting career implications are concerned, I suppose I should be attending too. In the past, my excuse was that ITA was an invitation-only conference, both for attendance and for paper presentation, and I had never been invited. Maybe my impression was wrong, or maybe this year it changed: it seems like the plebs can register, but papers are still by invitation.

Is there a good reason why ITA is invitation-only? It's clearly not intended to be a small gathering, and never has been: the participants page for the original workshop lists over 200 researchers. Perhaps the organizers are trying to spare themselves the effort of organizing reviews, but the IT community is reasonably self-selecting, and the Allerton organizers allow unsolicited papers.

On the other hand, I can think of three good reasons why ITA should not be invitation-only. First, I'm not sure ITA should be described as a "workshop" (which suggests "conference") if the purpose for non-invitees is to hear pre-selected lectures; that strikes me as more of a "school". Second, it reinforces the stereotype of ITSoc as a "clubby" society; as an information theorist, you are either one of the inner circle who gets an invite, one of the outer circle who doesn't get an invite (but who goes anyway to hang around with the cool kids), or a nobody. Third, ITA is killing CISS: at last year's conference in Baltimore, the decline in attendance was stark compared to past years; since CISS takes submissions, this removes an opportunity for the less-well-known to publish. Consider the impact of that on cross-disciplinary work.

On the other hand, maybe ITSoc is sending a message that there's no such thing as a less-well-known information theorist?

Social Media and the Modern Day Classroom

2010-02-04T17:00:00.000-05:00

(I was invited to speak at Social Media Week on Social Media and the Modern Day Classroom. These are my remarks.)

I was hired at York University in 2006, and -- like most new faculty -- I was full of ideas and energy for transforming undergraduate teaching as we know it. At the time I was also starting to play around with blogs, so it seemed natural to try to bring the blog into the classroom.

My idea was to replace the standard course website with a blog: the unchanging details of the course (schedule, location, etc.) could still be hosted in some static place, but the blog would communicate the day-to-day details of the course. My first attempt, still online, can be found here. It was an incredible success: the blog formed the core of a vibrant community, which allowed the students to communicate both with me and among themselves. One student was even inspired to start his own blog, transcribing my course notes after each lecture. Not all of the feedback was positive, of course; I kept the commenting system open to allow anonymous comments, and students took full advantage to express their frustrations and problems. But this was also a huge positive -- too often, the professor gets a sterile view of how the students are doing, because nobody is brave enough to speak up.

So began my efforts to integrate social media into the classroom. In addition to blogs, I've experimented with three other social media sites: facebook, YouTube, and Twitter. Each has its own unique features. Facebook and Twitter are quite similar: in Facebook, my approach is to create groups for each class; in Twitter, I create a class twitter account which students follow (and vice versa). These sites are, I guess you could say, "democratic" (although a less charitable word would be "anarchic") -- unlike a blog, there is far less room for the instructor to direct the conversation, and it shows. YouTube, though it re-creates the experience of listening to a lecture, turns out to be surprisingly passive: again unlike blogs, my students are reluctant to use YouTube's commenting features to leave feedback, which takes away from the sense of community (then again, perhaps this is not so surprising, since YouTube is comparable to the highly passive activity of watching television). In comparison, I would argue that the blog is as close as one can get in social media to the lecture: a blog post sets out a particular thesis, and permits an organized discussion on that thesis.

It's worth considering whether social media can replace the university classroom, and I'm going to cop out by answering "yes and no". For one thing, social media is unlikely to replace the small undergraduate class. To form a community based on social media, you need a critical mass highly committed "community builders", who are willing to jump in and participate in whatever media is in use -- be it a blog, facebook, or Twitter. In small classes, there simply aren't enough people to form a critical mass, so anyone trying to participate is left to feel awkward. However, for larger classes, social media does indeed pose an alternative to the traditional classroom order. We have already noticed the trend towards distance learning, so students are already willing to miss out on the impersonal 200-student lecture, even without social media tools at their disposal. A well-thought-out social media strategy, coupled with a high-reputation distance-learning program, could indeed recreate much of the classroom experience, and pose a viable alternative (or threat?) to the traditional university experience. The comparison between traditional universities and traditional media is chillingly apt.

Not because it is easy, but because it is hard

2010-02-01T20:05:00.004-05:00

Today President Obama announced that the United States would not be returning people to the moon, or sending them anywhere else beyond Earth orbit. Now, people can certainly argue that robots are more efficient, and spaceflight is expensive, and blah blah blah. But I think I'll leave the case for manned space flight to the man who proposed the lunar project in the first place.

Formspring

2010-01-26T10:36:00.003-05:00

Continuing my tradition of screwing around with "social media" sites, I'm now andreweckford on formspring. Here's Gawker's take on this question-and-answer service, although my goals for the site are more professional than personal.

Speaking of YouTube

2010-01-22T20:05:00.003-05:00

White, middle-aged men should never rap -- they just end up looking like dorks. But I'll make an exception for this guy, 'cause if you're going to dork it up, you may as well go all the way. That mofo is ghost riding a Prius, yo:

[@IBMResearch via Gizmodo]

Disappointing arXiv paper title of the week

2010-01-21T12:58:00.005-05:00

When I saw the paper Real Interference Alignment, I initially read the title as "Real-world interference alignment", but what they actually mean is "Real-number interference alignment".

This is not to criticize the paper -- Amir Khandani and his group do great work, and this paper looks nice. But I'm still hoping for a paper explaining how to achieve interference alignment under practical, real-world assumptions.

I won't be seeing you in Austin

2010-01-14T14:51:00.003-05:00

For a variety of personal reasons (not the least of which is currently nine months old), I didn't get my act together to submit a paper to ISIT this year. I had a couple that I was working on, but they didn't come together in time. So I'll be missing my first ISIT since 2003 in Yokohama.

St. Petersburg is going to be a blast, though. Assuming the LHC hasn't destroyed the earth by then.

Information theory for the YouTube masses

2010-01-11T19:31:00.005-05:00

I'm teaching a graduate-level course on information theory this term. As I've done with a couple of my other courses, I'm going to make video recordings of the lectures and post them online.

The course YouTube channel is here. As I've also done elsewhere, I'm organizing the video and course announcements on a blog, which is here. I just posted the first lecture -- this is still kind of experimental, so comments would be appreciated.

As far as I know, this will be YouTube's first full course on information theory. However, while searching for other IT-related material, I ran across this gem. Plus a slew of raging creationists and evolutionists, each using information theory to somehow try to prove their point.

Happy new year

2010-01-04T12:44:00.002-05:00

I'm back at work for the first day after the holidays. Posting has been light for the past while, but I plan to be more disciplined about this blog in the new year. My goal for 2010: 52 blog posts, and at least two per calendar month.

Conference fraud actually happens

2009-11-29T08:35:00.008-05:00

Some time ago, I wrote about the possibility of fraud in academic publishing. We all get spam promoting new academic conferences, but this spam seems to enjoy a presumption of legitimacy. As I argued, this presumption could be exploited.

It appears I was right to be wary. The Scientist has uncovered two recent cases of fraudulent conferences.

The first case unfolded largely as I thought: a general invitation went out for the "1st International Cardiology Conference" in Shanghai; invitees were directed to a legit-looking web page (since removed), and registration fees were charged. Not only was the conference bogus, but registrants' credit card numbers were used for further fraudulent charges, totaling $2000 in one case.

The second case is more reminiscent of the traditional Nigerian scam: a prominent researcher was invited to appear at the "Seventh Annual International Global combine Conference on Global Economy and Human Welfare", with free airfare and accommodation; the invitee had only to provide certain personal information. The recipient of this message was wary enough to check with the supposed sponsoring organizations and venue, none of whom had ever heard of the conference.

I guess I'll be sticking with the usual IEEE conferences for the time being.

[via BoingBoing; h/t @adhamilton.]

He was the learner, but now he is the master.

2009-11-18T15:16:00.002-05:00

My first master's student just successfully defended his thesis. Congratulations, Nariman!

Running with memes

2009-11-18T09:42:00.010-05:00

Let's go with this.

The answers are: yes, more than lawyers but less than doctors, sometimes, no, depends on when you ask, better than baristas but worse than doctors and lawyers, mostly, I sure am, I sure am, and I hope I'm not.

In a computer science and engineering department, this is the kind of question that can start riots.

Are they indeed? (Sadly, if you follow this search it takes you to pages of lame "computers are like women/men" jokes).

Time-travelling censorship

2009-11-06T09:35:00.005-05:00

I read this a while ago and kind of laughed. But then this happened. I mean, really? A piece of baguette dropped perfectly in the wrong place? Somewhere, Douglas Adams is smiling.

The future of tenure

2009-11-05T16:16:00.004-05:00

Picking up where Anand left off a couple of weeks ago, the current issue of Academic Matters features a three-way debate on the future of tenure: Michiel Horn argues in favor of tenure, while Michael Bliss and Mark Kingwell argue against. Kingwell's article is, verbatim, the one from Inside Higher Ed.

Bliss's article comes across as a rant, while Kingwell continues his tendency of writing articles that are more attention-grabbing than well-thought-out.

However, I'm struck by Horn's argument: abolition of tenure could lead to the worst of both worlds (worst, at least from the tenure critics' point of view): hard-to-fire professors who are never required to demonstrate their productivity. Horn puts tenure in historical context, showing that it emerged in the 1960s as a financial perk, and a codification of the long-established practice of respecting the academic freedom of professors. Without tenure, the system could well revert to the historical norm of automatic contract renewals -- but this time without the need for pre-tenured faculty members to demonstrate high achievement.

A university without tenure might resemble the public school system, where it is rare for teachers to be dismissed for poor performance. I suspect this is not what Bliss and Kingwell have in mind.

Nanonet 2009: Video

2009-10-26T12:35:00.005-04:00

I presented a paper at Nanonet 2009 in lovely Luzern. I wrote about the paper earlier.

Here's a video of my conference presentation (playlist):

Paper at Nanonet 2009

2009-10-26T12:26:00.003-04:00

My paper at Nanonet 2009:

A. W. Eckford, “Timing information rates for active transport molecular communication,” in Proc. 4th International ICST Conference on Nano-Networks, Luzern, Switzerland, pp. 24-28, 2009.

PDF of the paper is here.

The paper is a first stab at molecular communication in a microfluidic system, like a lab-on-chip. This is one environment in which molecular communication would be commercially feasible: it turns out to be very hard to integrate electronics on these devices, so as much as possible needs to be done in liquid and chemistry. In the paper, I only look at timing information, and compute some achievable information rates. Plenty more work to follow.

Who can stop old RMC? (again)

2009-10-06T09:37:00.003-04:00

My glorious alma mater strikes again. Former Royal Military College physics professor Willard S. Boyle wins a share of the 2009 Nobel Prize in Physics, for his work on charge-coupled devices. (He did his Nobel work at Bell Labs.)

Canada enters 21st century

2009-10-05T20:14:00.002-04:00

Ever since Rogers bought Fido five years ago, anybody wanting a GSM phone -- which is to say, anybody wanting a phone that would actually work in other countries -- was stuck with a single choice. That's about to change.

The right to remain silent

2009-10-04T09:26:00.000-04:00

We've been warned about living our electronic lives online, as what we say can be used against us later. However, here's the first example I've heard where an adverse faculty hiring decision may have been based, in part, on the candidate's blog.

This week, Mihai Pătraşcu -- a theoretical computer scientist at AT&T Labs -- wrote a blog post in which he discussed his recent academic job search. I can't comment on his work, being well outside my expertise; but many people in his field (like Mitzenmacher, where I found the link) clearly think of him as talented. Pătraşcu's post is (in part) about his interview at UCSD -- they gave him an offer, but only after their first choice turned them down. On the basis of being ranked second, Pătraşcu declined the offer.

Pătraşcu mentions that, on the whole, his job search went less well than he expected: he didn't get interviews at many of the "top places" he applied. But down in the comments, we read:

I was on a hiring committee in one of these schools that decided not to interview you. Although I hesitated to post this comment, I think what I have to say will be helpful to your career. The reason we decided against further considering your case was because of your reputation as a very difficult, arrogant, and opinionated person. We even read your blog and found many posts that confirmed this reputation.

Of course, this anonymous commenter could be anybody. But the tone of Pătraşcu's post is certainly abrasive, so it is quite plausible that a hiring committee would have looked at similar posts and decided not to interview him.

I'm ambivalent about this. On the one hand, if his blog is representative, this guy wouldn't be fun to have as a colleague. On the other hand, I'm worried about the chilling effect -- it would be too bad if young researchers stopped blogging for fear of damaging their careers.

Giving up, and giving in

2009-10-02T19:47:00.004-04:00

So I'm @andreweckford on Twitter now. We'll see how this goes.