Monday, November 10, 2014

Twin Paradox: The Olfacto-mathematical Delusions of the Varshney Brothers



Like movie stars wearing prop glasses to make themselves look smart, pop media outlets this week splashed themselves in eau de science, trying to out-cool each other with superficial coverage of a new paper in olfactory signal theory. [That’s a thing?—Ed.] [It is now!]

What had the click-bait crowd all aquiver was an abstruse theoretical paper by a pair of electrical engineers who happen to be twin brothers. Kush and Lav Varshney both received undergraduate degrees from Cornell in 2004 and doctoral degrees from MIT in 2010. They’re obv very smart and good at math. (They are also the third generation of their family to become electrical engineers.) Kush is now at IBM’s Watson Research Center in New York, while Lav is at the University of Illinois, Urbana-Champaign.

Their new brainstorm-in-common has the dry title “Olfactory signals and systems.” As befits a theoretical engineering paper, it contains a lot of equations; like Einstein, they leave the actual data-gathering and experimental validation to others. That’s cool—everyone enjoys a good theory, especially if it makes outrageous claims. And there’s been a market opportunity since September when Luca “Vibes” Turin got peevish and flounced off the Twitter stage.

The claims made by the Varshney bros are indeed click-worthy. They claim to set out the theoretical basis for “odor cancellation” technology as well as “food steganography.” [Stega-whaa?—Ed.] [Go look it up yourself.]

While mere mortals have struggled for centuries to characterize and quantify smells, and to analyze their chemical basis, the V twins achieve their theoretical breakthrough in three deft strokes of matrix algebra. First, assume an array consisting of all the physiochemical properties of every odor molecule (molecular weight, functional groups, yadda). Next, assume an array consisting of all the perceptual vectors in odor descriptor space (ratings of strength, pleasantness, green, floral, yadda). [One is inevitably reminded of the old punch line about spherical cows in a vacuum.] Finally, produce a matrix that maps molecules from the chemical to the perceptual space. Now all you have to do is solve for



and, voilĂ ! You can now predict the smell of any molecule from its physical properties or vice versa.

The V bros helpfully point out that the problem “is convex and can be solved by interior point methods and a variant of Nesterov’s smooth method.” [Of course. Why didn’t I think of that?—Ed.]

Lest it be said that electrical engineers lack a sense of humor, note that the physicochemical array is denoted Rk and the perceptual array is Rl. Get it? Those guys are a laff riot.

The first big claim made by the Varshneyii concerns “active odor cancellation,” an awesome outcome in which their system identifies a small number of unobjectionable compounds that will negate a particular nuisance odor or stink—knock it right out of your nose! At least that’s what normal people would think, given the analogy with active noise control. Noise cancellation systems deploy a sound wave with an inverted phase to that of the noise wave; the perceptual result is no sound at all.

Unfortunately, that’s not exactly what Tweedle-Kush and Tweedle-Lav are selling. Instead, their matrix solution spits out a set of odors that, combined with the stink at issue, produces a standard jumble smell known as “olfactory white.” True, olfactory white may be less obnoxious than the malodor in question, but it is still a smell, not the absence of smell. This is not the equivalent of active noise cancellation. When I use noise cancellation headphones on an airplane, the steady roar of the engines disappears. It is not drowned out by the addition of white noise.

Oh, well. On to the second claim: food steganography. In this application, the matrix solution provides a set of other odors that effectively masks the objectionable aroma of a particular food item. The Varsity Varshney say “nuclear norm-regularized multivariate linear regression.” I say, think bacon-wrapped chicken livers.

I’m not sure exactly what it is about olfaction as a scientific field that attracts charlatans, nitwits, poseurs, and grandiose humbugs. But I’m sure someone will come up with an equation for it.

The paper discussed here is “Olfactory signals and systems,” by Kush R. Varshney and Lav R. Varshney, published online October 17, 2014. It can be downloaded here.

No comments: