bekolay.org

In what is a more traditional (read: uninteresting) blog post, today I will let you know what I've been up to, then rant.

Cracked.com's Michael Swaim. He's pretty funny, I guess.

With a project deadline quickly approaching, I've been more productive than ever. Soon, I will even start working on my project.

If you've watched The Hangover, there's about an 80% chance that you liked it. I really didn't, so I spent entirely too long making fun of it over on Cracked.com.

Pictured: The Computational Neuroscience Research Group.
Not pictured: Any girls, cripes.

Back when I was a journalist, I would spend far too much time researching each article, only to hurriedly write up what I'd learned minutes before (or perhaps after) my deadline. As a researcher, I find that as I wait for my experiments to finish running, I feel the need to write (and Photoshop (see above)).

A lot of researchers find publishing to be a hassle -- an unwelcome interruption of their research. Personally, I think it's more of a perfection thing: they want to publish the best results ever, and become god-like in their field. I can definitely understand that, and perhaps I will feel the same when the time comes to submit abstracts to academic journals, but in my opinion there is still merit (academic merit even) in writing as much as you can.

I remember reading a quote at some point that the further you go in academics, the more important it is to be a good communicator, both in print and in presentations.

Then today I read a this fantastic article about Grigory Perelman, the man who proved the Poincaré conjecture. It is a super badass story, worth a read if you have time.

One of the things that struck me about the article is that he really could give a shit about communicating -- in fact, he has not communicated with anyone other than his mother in quite some time. A geometer (dude that is all about geometry), when talking about Perelman, says, "the ideal scientist does science and cares about nothing else. He wants to live this ideal."

For better or worse, I can't be that kind of scientist.

One thing I definitely agree with him about is academic honesty issues. Perelman has now essentially quit mathematics, saying, "It is not people who break ethical standards who are regarded as aliens. It is people like me who are isolated."

I obviously don't have the experience that Perelman has, but something that has been bugging me lately is how so few of the papers that I'm reading publish their source code.

I do recognize that one more clearly understands a problem if they are forced to implement the solution themselves, but this is hardly a justification for the majority of code written in academia to be kept secret within each lab.

I think that publishing source code should be the norm, especially in mathematics and computer science. It should be seen as suspect when source code is not provided. And that source code should be made available to other researchers to build on. Software packages like ACT-R and OpenCV have become standard in their field not because they are the best, but because they are the best available.

Whatever scares people away from publishing source code, these projects -- as well as decades of open source software development -- should be strong evidence that sharing is good.

When I talked about this with a post-doc in my lab, Terry Stewart, he shared my concerns and even recounted a number of studies that have tried and failed to replicate the results of published results in a number of fields. Further details can be found in his PhD thesis (skip to chapter 5).

He showed me a website he helped create, which aimed to be a central repository for computational models associated with scientific papers. Unfortunately, it is nearly barren of content, the last contribution being from over a year ago.

The idea is sound, but I really think that the key to having a system like this work is to get the cooperation of big journals; if it becomes standard practice to submit your source code when submitting to ACM or IEEE, then smaller journals are sure to follow (especially if the software to link articles with code is well developed).

Unfortunately, due to the inherent fuddy-duddiness of old, well established journals, I am not confident that this will happen. Perhaps someone needs to start a site like arXiv.org and start a revolution that way. I think with a lot of young researchers raised on open source software just starting to start making academic contributions, it could work.

Perelman, by the way, published his Poincaré proof on arXiv.org, meaning it was immediately available to everyone for free.

Since I'm on an academic rant, I figure now's as good a time as any to show this video produced by the lab I'm a part of. It was just shown at NIPS, and demos a few simulations in our neural modelling software nengo. To really understand what it's doing would take some time to explain (I don't quite understand it all myself yet) but it's still worth giving up five minutes to check out. Plus you get to see where I work.

Rest assured, as I find out more about what nengo's doing, I'll post details (time permitting).

Oh, also, I found a super kickass fixed-width font today called Inconsolata. It's free, so do yourself a favour and start using it today for all your fixed-width needs!