Thread: What do you work at???

Originally Posted by wdhall

I'm a mathematician, doing algorithm research & development on a continuous blood glucose (+ other analyte) monitor for the ICU (hospital intensive care unit). If it weren't for the facts of heavy medication, organ failure, and severe blood loss / fluid replacement therapy, this would be a walk in the park. Well, maybe a park that required ropes & pitons to negotiate, but a park nonetheless. With those other factors, it's a bit trickier.

Easily the most challenging and rewarding job I've had in the ~30 years since my Ph.D.

pretty cool, what kind of algorithms do you work with?

Guess my last post didn't take. I'm a public guardian. I work with the cognitively impaired who have been determined by the courts to be unable to make informed decisions. I make decisions about their medial needs and in certain other areas.

I am a HATTER (hat maker) for the past 14 years

Originally Posted by str8tshaver

I am a HATTER (hat maker) for the past 14 years

You don't happen to be insane by chance, even slightly deranged? I had heard of a fellow known as the Mad Hatter and it would be a remarkable coincidence if you turned out to be he.

Originally Posted by str8tshaver

I am a HATTER (hat maker) for the past 14 years

What kind of hats do you make?

Freelance RIA developer.

Originally Posted by fatpanda

pretty cool, what kind of algorithms do you work with?

Currently, I'm using a version of linear regression called PLS (it's pretty much the popular favorite in the chemometrics world, probably because it's so much like ordinary regression that people are comfortable with the technique, and have good intuitions about what works and what doesn't); in the past, I developed [and then used] a non-regressive technique (HLA) based on a '98 Ph.D. thesis out of the MIT Physics Department; however, some spectacularly uncooperative data together with requirements of expediency nudged us towards PLS, where other folks in my team could contribute meaningfully. I'll probably get back to HLA eventually, but it'll take more time & effort than I have at the present.

Due to the hospital environment we're targeting, PLS requires a number of tweaks to do the job. Mainly, the problem is that ICU cases are by their very nature, custom jobs: no two patients have the same situation, doctors are pumping this and that into the patient to keep him alive, then to keep him stable while they figure out what is going on. The problem to me is that everything they do has an impact on the blood, specifically on the spectrum that I get to see. The blood samples one obtains can be markedly distinct form anything your training set wants to see, and regression can (and does) fail catastrophically.

To mitigate this effect, I have developed techniques to identify [probabilistically] the interferents most likely to be the cause of any spectral discrepancies (relative to the training set), and then to take a list of presumed interferent spectra and re-tune PLS to be relatively insensitive to their presence.

The interferent identification uses a Bayesian technique to estimate relative likelihoods for the various interferents from my library; the re-tuning is essentially an expansion of the training set to include the identified interferents as part of the background noise. The Bayesian technique works well enough for a modest number (< 5) of simultaneous interferents, while the retuning works exceptionally well for upwards of 15 simultaneous interferents.

That's enough of me rambling & bragging.

Dale

Let me see if I understand what you just said- you use certain mathematical techniques in order to better understand the effects of certain medications and environmental factors in order to have an effective real time blood analysis monitor?

Started work with TELSTRA RESEARCH LABORATORIES at 15 years old, I did a Fitting & Maching apprenticeship, then Tool Making (plastic injection moulding) Got a degree in Mechanical Engineering. Worked there for 30 years until the government closed it down and made me reduntant. ( great party while it lasted) Past 7 years I have been a postal delivery officer (postie) Get to ride a motorbike and be home by lunchtime .
Only 4 years to retirement

Originally Posted by Quick Orange

Let me see if I understand what you just said- you use certain mathematical techniques in order to better understand the effects of certain medications and environmental factors in order to have an effective real time blood analysis monitor?

In a nutshell, yes.

Of course, as anyone who has attempted anything non-trivial will tell you, the devil lives in the details. That is, to get anywhere with this sort of problem, it has to be cut way down to manageable chunks: what specific analytes do you detect, what substances can possibly be in the way of your seeing the thing you're after, how big are they, what does normal blood look like anyway, and so forth. Some of the answers come down to doing some computations, others require folks to do a bunch of literature research, still others involve getting blood, lots of it, and analyzing it every which way that makes sense.

So, the real answer is a bit more nuanced than "yes". It's more "yes, but"... where you get the small print: you can't get a complete list of ingredients (only analytes in some short list), you can't be absolutely immune from the effects of absolutely any combination of contaminants at absolutely any concentrations (only a list where we've: developed a high-fidelity spectrum, and have some clue about how much of that stuff can possibly be present; and only at blood levels up to some moderate-sized multiple of the maximum level observed in our literature search). Besides those limitations, if you have sufficiently many high-level contaminants, all bets are off, and for good reason: if you give me at least as many contaminants as I have free variables in the regression equations, I can produce a combination of those contaminants that will reproduce whatever other spectrum you care to pick, including the very thing you thought you were looking for.

So, you can't always get what you want. But if you try sometimes...