Competition crowdsources blisteringly-fast software If you want a massive improvement in the software you use, the cheapest way to get it is to host a competition on TopCoder. That seems to be at least one of the discoveries made when a group of research biologists staged a competition on the MIT-operated site. A two-week contest with regular prizes of $US500 ended up costing …
An algorithm can not be patented.
Of course, solving a problem that potentially you could sell for thousands more may be considered stupid, but is does show you skills off and perhaps some of these people will be employed by the supposed market leaders to improve their software.
People compete on TC largely for the sheer fun of it. Most problems have no prizes at all so the potential of a prize (most members are not in the West) and some recognition are just juicy bonuses.
This is proper all-night-coding-session-and-then-going-to-school nerdery, real stereotype stuff other coders here will no doubt identify with - I know I do (I used to compete but ran out of time).
Or what's almost as bad is that they assume they are merely competent in another field and fail to reach even those levels.
The biggest example that springs to mind is the numbers of non-specialists who think they're competent in statistics when in reality all they can do is (mis)apply formulae and algorithms whose derivation they don't understand.
The competition is proof that they thought that someone else could achieve what they admittedly failed to do
You'll also notice that what you quoted is a quote from the original article, not a quote from the scientist. The original article continues with a real quote from the scientist:
“This is a proof-of-concept demonstration that we can bring people together not only from different schools and different disciplines, but from entirely different economic sectors, to solve problems that are bigger than one person, department or institution,”
It wasn't a "firm" that developed the benchmark software (MegaBLAST). MegaBLAST is a variant of the standard BLAST algorithm, tuned for finding close matches for long sequences (as an alternative to running multiple BLAST searches). It's maintained by the NIH.
The BLAST algorithm was published in 1990. Predictive-matching algorithms have evolved a bit in the nearly-a-quarter-fucking-century since then - thanks in no small part to BLAST, which was very influential. The constraints on computing resources are a little different, too.
You might also try getting your facts right. Arnaout's custom solution, the other benchmark for the project, was already an order of magnitude faster than MegaBLAST; the best Topcoder solution improved that by another order of magnitude. That's certainly very good, but comparing Topcoder's result only with MegaBLAST is pure sensationalism. Also, the Topcoder project cost the team $6000, not $500. It doesn't matter if the final stage only cost $500 - the overall cost was $6000. So they got a 5x performance improvement for free, and then a 25x performance improvement on top of that for $6K. (I don't know where the "970 times faster" comes from - the case-study link above claims the overall reduction in wall-clock time was from 2000 seconds to "just over 16 seconds".)
But, hey, why not post some idiotic, insulting comment about people you don't know and work you don't understand? This is the Reg, after all.
Similarly, I have to wonder whether the folks posting about the commercial potential of this work have bothered to look at the algorithms used by the successful entrants. Were they novel, or were they just an application of algorithms already in the literature to a new domain? (I've looked for more detail but haven't found it yet, nor have I found the "codes ... available for free downloads" mentioned above. Though I haven't tried that hard, to be honest.)
Generally, from having taken part in these contests before, they are designed so you can't just find an algorithm and use it. You can find ideas from certain areas of CS but then have to figure out how to use them. Algorithms designed in these contests tend to be HUGELY tied to the very specific way input data is provided... e.g. the random distribution used in generating test cases will affect the solutions posted.
The top guys really are VERY smart.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
