Watson is used here the same way people use Hoover and Kleenex. I thought the article would be about IBM beginning to refit their question/answering inference system for aiding in diagnosis at hospitals. But it is not.
The article is about researchers monitoring large streams of data and using machine learning to help in detecting whether an infant in the ICU has an infection. The data streaming tech is from IBM though and based on what was used in Watson, so there is that link.
Interesting work nonetheless and hopefully an echo of things to come. Perhaps those working in the analysis and prediction on massive streams of data part of HFT could reapply their skills here in a way that could literally save lives.
Oh I did not know that. I guess I was tricked by the wording in this sentence:
"Artemis is built on an analytics platform called InfoSphere Streams that, like Watson, emerged from IBM research into ways that software can make decisions on the spot using data arriving at a high speed from many different sources".
I thought they started from the same origin and then split focuses. Thanks for the correction.
Good to see that Watson is making the transition to a useful expert system/aid-to-humans quickly; and interesting to see that David Ferrucci's and Chris Welty's comments about adapting Watson from Jeopardy use to medical use came to fruition only 2 months after they made them during a showing at RPI of the Jeopardy episodes in which Watson competed.
Fair enough. I'm reluctant to say too much since I actually work on Streams (hence my authoritarian corrections), but I'm a research staff member, not a marketing person.
I suppose I do feel comfortable giving my personal take on it, and why I find working on Streams exciting: we're designing a language and runtime system for a new programming model. Not just a new language (which we do have), but a new programming model. The way you write programs changes when you have essentially infinite streams of data coming through your system - you have to think in terms of operators that transform or filter individual pieces, and you have to keep in mind its inherent distributed nature. That is, you may write a chain of operators to process your data, but each operator can run in parallel with the others, even though you will probably design your application thinking of one particular piece of data marching through the system. The runtime system itself is, of course, distributed and fast.
To be a bit blunt, that article with a heavily misleading title totally disappointed me. I was expecting to glean some insights there on how they applied certain kind of technologies into a vertical market -- nothing but a few marketing pitch.
Your post got me even more confused: what exactly new programming models? I understand there may be an influx of sensor data coming in and your apps have to be able to cope with them --- do specific data analysis, maybe some kinds of expert systems, and generate possible suggestions. But how is that different from existing systems, say, nuclear station operation system, high-frequency trading systems, to name a few.
Those systems exist, but they are ad-hoc. That is, people implement their own in-house solutions as needed. We're trying to provide a language and runtime that allows developers to only focus on writing streaming applications - they don't need to worry about implementing their own infrastructure or way to describe their applications.
Infections and acute illnesses in infants in general are scary for a number of reasons. These guys can go bad very quickly. They can't tell you if and where they hurt, so it's a little like veterinary medicine.
In training, very subtle clues like a slightly elevated heart rate or (in particular) respiratory rate didn't initially alarm me because in adults these kinds of isolated findings mean practically nothing. Very experienced peds nurses would occasionally take a look at a baby I'd just examined and wasn't worried about...and would worry. I learned quickly to trust what seemed like intuition. We'd start the wheels in motion for more vigorous intervention.
These nurses (and the docs who then benefit from their experience) were applying a lot of (sometimes unwritten, if not unconscious) heuristics in these evaluations. A system doing this rigorously, objectively, and routinely is a good application of such software.