127 karma · joined January 25, 2016
they also use this as a disingenuous/insincere way of rejecting founders (we don't want to invest, but you're probably one of the other unicorns we didn't invest in! check out out antiportfolio!)
we've had a wonderful positive response to this article, and people are overwhelmingly excited and motivated to try and exercise more because of it (even at relatively lower intensities or volumes -- every little helps). even if we are extremely cynical and say theres zero benefit of exercise for mental health, then people will still be getting known improvements regarding diabetes, stroke, cardiovascular disease. Nobody is advocating for exercise at the expense of alternative treatments like meds or therapy, this really is just a (potentially large) extra benefit.
2- nope, and actually my entire research agenda has been around helping people find a treatment that is most likely to work for them specifically! (we also founded a company to bring the research to market- springhealth.com)
3- that is beyond the scope of the paper, and we didnt use the word hospitalized once
4- yep, and thats a good example of one of the pitfalls of self-reported surveys. in general, i am a bit less cynical and in our experience, most people aren't filling out surveys to intentionally lie or sabotage.
hope this helps!
Also sorry about the paywall - you can get the article either through sci-hub here http://sci-hub.tw/10.1016/S2215-0366(18)30227-X or you can email me and i'll send you a copy direct!
For example, here's a paper from the 1980's also predicting when a patient will die, also using a couple of thousands of patients' data: http://europepmc.org/abstract/med/3816253
And, Google's paper wasn't published in Nature, it was in a new open access journal owned by Nature. In academia, that is like the difference between a really fancy porsche, and a really cheap volkswagen (two very different things, both owned by the same company).
i wonder if this would have been improved by being clearer about the motivation. the authors frame it as though the ~60ms penalty for using the LSTM for prediction is a huge burden, and i can imagine situations where it is. however, it seems like if this is the case, we need some real life/"scaled out" examples of how this solution would work in practice. e.g. how long does the decision logic take to execute (maybe 5ms?); what proportion of the time will you have to run the LSTM after the BoW model anyway? note that those instances you are now worse off than just running the LSTM in the first place (total time = BoW time + decision time + LSTM time). once you have all these you can run the math and know (on average) how much time you'll actually save, and how much performance you sacrifice
IMO theres an awkward space between small data and big data where it isn't really worth spending a long time to treat it like a real "big data" problem, and the x1 instance gives you an easy-out.