11 comments

[ 5.3 ms ] story [ 30.7 ms ] thread
Sounds like something to build as part of your tricorder and get paid twice :oP http://www.reghardware.com/2011/05/12/x_prize_foundation_tri...

Also, that doesn't sound like a lot of money to transfer intellectual property rights for that. I guess they have patents on the hardware side so the algorithm is not easily monetisable by itself.

If you view this as a freelance project instead, it is a fixed price $45,000 project where you only get paid if you 1-beat the best result of a team of professionals who have been doing this, 2-are the first to do so, and 3-achieve something that may or may not be possible.
The 2nd chapter in Head First Java has a competition for an Aero chair between a procedural programmer and an OO programmer. The narrative describes how the OO programmer dominates the procedural programmer throughout the competition. So, who won the chair? Well, according to the book it was the 3rd programmer neither knew was offered the contract.

Seriously, I guess this would be fun if you and a few buddies were bored and in school. Otherwise, no thanks.

(comment deleted)
45,000 $CAD, not $USD
Right, so more than $45,000.

1 Canadian dollar = 1.0324 US dollars $45,000 Canadian dollars = $46,458.90 US dollars

It's been a few months since a US dollar was worth more than a Canadian dollar.

they want 95% classification done using 16KHz sampling of "passive electromagnetic emission" (i.e. basically noise).

Good luck classifying between me and Pavarotti only by listening to our breath over the phone.

For reference, the wavelength of an 8 KiHz electromagnetic wave (the Nyquist frequency at their sampling rate) is over 36 km. (Edit: at the speed of light in vacuum, that is.)

Perhaps I'm misunderstanding something, but this would seem to limit the spatial resolution that could be achieved by sampling a single stream of passive voltages at that rate, and I would expect differences between materials to mostly be on the microscale. But they don't say what other constraints might be in effect or what the materials or environment are, either, so…

If they're currently getting 79% accuracy, there must be some significant signal there. No idea if 95% is achievable, but unless they're already using the absolute optimal approach, something at least into the 80s should be possible.