3 ms·
Some time ago I wrote about academia vs industry in the context of data science: In academia, you are allowed to cherry-pick an artificial problem and work on
by stared 9y ago
Some time ago I wrote about academia vs industry in the context of data science:
In academia, you are allowed to cherry-pick an artificial problem and work on it for 2 years. The result needs to be novel, and you need to research previous and similar solutions. The solution needs to be perfect, even if not on time.
In industry, you should solve a given problem end-to-end. Things need to work, and there is little difference if it is based on an academic paper, usage of an existing library, your own code or an impromptu hack. The solution needs to be on time, even if just good enough and based on shady and poorly understood assumptions.
So, contrary to its name, data science is rarely science. That is, in data science the emphasis is on practical results (like in engineering) - not proofs, mathematical purity or rigour characteristic to academic science.
(from http://p.migdal.pl/2016/03/15/data-science-intro-for-math-phys-background.html#priorities http://p.migdal.pl/2016/03/15/data-science-intro-for-math-ph...)
- denzil_correa 9y ago> So, contrary to its name, data science is rarely science. That is, in data science the emphasis is on practical results (like in engineering) - not proofs, mathematical purity or rigour characteristic to academic science. Science in a nutshell is to take a hypothesis, test it rigorously and validate/invalidate the hypothesis based on the results. The results of this experiment could be extremely practical. Data science that follows the scientific experimental design, does science. The one that doesn't is equivalent to reproduction of past experimental results OR putting a blindfold on your eyes and throwing darts around until it hits the bull's eye.
- stared 9y agoI disagree with this common, yet faulty, explanation. (IMHO coming from people who have never worked in science and for whom everything R&D looks like "science!".) Building engines or cars (involving a lot of tests!) is considered engineering. And as any serious engineering it's not purely random trails&errors. A/B tests are not a sufficient criterion to make something science. And data science (in business) is focused on solving practical problems (which may or may not be research) rather than open-ended research (which may or may not be practical). Source: I worked in academia. I now work in data science. You?
- denzil_correa 9y agoI have a Ph.D in Computer Science and I have worked across research labs in the industry too so I very much understand where you come from. That said, I don't see the disagreement between us and what part of my explanation you found faulty.
- stared 9y agoSorry for being combative. It comes from two things: - giving talks at universities (then I admit that "data science" is more engineering, compared to academic science) - people who had only experience with software development and then any research-like stuff (or simple mathematics) suddenly becomes "science!"