Slashdot Mirror
Machine Learning Expert Michael Jordan On the Delusions of Big Data
First time accepted submitter agent elevator writes In a wide-ranging interview at IEEE Spectrum, Michael I. Jordan skewers a bunch of sacred cows, basically saying that: The overeager adoption of big data is likely to result in catastrophes of analysis comparable to a national epidemic of collapsing bridges. Hardware designers creating chips based on the human brain are engaged in a faith-based undertaking likely to prove a fool's errand; and despite recent claims to the contrary, we are no further along with computer vision than we were with physics when Isaac Newton sat under his apple tree.
There's plenty of reasons I can think of why I'd prefer image recognition on my phone rather than the cloud. Privacy, for one. If you let FB tag your photos with the names of the people in it (after teaching it those names), what do you think happens to that data? You might not even want to share the photo or video stream with anyone... Another reason is that we still do not live in a world with ubiquitous and cheap mobile data. Travel abroad, and you'll find out quickly why cloud-based services like Waze aren't always a viable option.
If construction was anything like programming, an incorrectly fitted lock would bring down the entire building...
I think he underestimated the power of stupidity.
You can grant every reasonably well-off person in a country a device that gives them access to all scientific and engineering knowledge and a vast communications network - and half of them will use it to publish rambling arguments that the moon landing was fake, fossils are a hoax scientists made up to disprove the bible, autism is caused by vaccines and Obama is secretly a Kenyan Muslim Communist Atheist Black-Supremecist who hates America.
In reality, it's only used where it doesn't really matter.
That's patently false. It's used for industrial process control and things like that too. See for example the video I posted. To the manufacturers who use such a machine, it matters an awful lot.
OCR - mistakes are corrected by spellcheckers or humans afterwards.
I don't know how much you count this as "mattering". The IEEE has scanned and OCR'd their back catalogue of papers. I don't think they've been human checked due to the sheer volume. It's very useful to be able to get these online now.
Mail systems - sure, there are postcode errors, but they result in a slight delay, not a catastrophe of the system.
Well, it's not like humans are error free either. This is something people often forget. A national postal system is a very important thing, and CV is used to massively reduce the costs of being able to ship vast quantities of mail. Sure it makes mistakes, so do hand sorters. By an astonishing coincidence, I actually got a letter through my letterbox for my neighbour only yesterday.
Structure from motion - fair enough, but it's not "accurate" and most of that kind of work isn't to do with CV as much as actual laser measurements etc.
I'mn not sure what you mean by "not accurate". It has a scale ambiguity, for sure, and drifts, but so does any relative measurement system including lasers.
Photo stitching - I'd be hard pushed to see this as more of a toy. It's like a photoshop filter. Sure, it's useful, but we could live without it or do it manually.
Well, of course we could live without it. Turns out that humans can survive with nothing more than a pointed stick and a bit of animal fur. This means we could survive without almost everything around us.
Anyhow, I doubt you'd get remotely comparable results by hand. You have things like vignetting, exposure changes, radial distortion etc to contend with. It's very, very hard to get a seam-free stitch.
Number plate recognition - well-defined formats on tuned cameras aimed at the right point, and I guarantee there are still errors. The systems I've been sold in the past claim 95% accuracy at best. Like OCR, if the number plate is read slightly wrongly, there are fallbacks before you issue a fine to someone based on the image.
But all systems have errors. Humans are quite error prone, especially in really boring repetitive tasks. One thing I've noticed is that where humans are really really good, they're held up as a gold standard, when they're not, perfection is held up as a gold standard.
Face detection is a joke in terms of accuracy. If we're talking about biometric logon, it's still a joke. If we're talking about working out if there's a face in-shot, still a joke. And, again, not put to serious use.
Face detection (not face recognition) works "pretty well", I reckon. You can download an old, non-state of the art algorithm like Viola-Jones in OpenCV. It's pretty good on the whole. And anyway: define "serious". But yeah, biometrics is a joke. I never would claim otherweise.
QR scanners - that I'll give you. But it's more to do with old barcode technology that we had 20 years ago, and a very well defined (and very error-correcting) format.
No, the old tech was laser or LED based scanning. The current ones use computer vision to avoid those complex, mechanical systems to be able to do a pretty good job with ubiquitous off the shelf sensors. Also, a generic vision based one can read pretty much all formats from a single place.
Pick-and-place rarely relies on vision only. There's much better ways of making sure something is aligned that don't come down to CV (and, again, usually involve actually measuring rather than just looking).
Sure they use servos and stuff for positioning, but those little crosshair marks over the board are what they use to get the high accuracy. The problem with the cheap-ass Chinese machines for a few gr
SJW n. One who posts facts.