Follow Slashdot stories on Twitter


Forgot your password?

Slashdot videos: Now with more Slashdot!

  • View

  • Discuss

  • Share

We've improved Slashdot's video section; now you can view our video interviews, product close-ups and site visits with all the usual Slashdot options to comment, share, etc. No more walled garden! It's a work in progress -- we hope you'll check it out (Learn more about the recent updates).


Comment: Re:Do they actually work well now? (Score 1) 45

by lorinc (#49136361) Attached to: The Believers: Behind the Rise of Neural Nets

Last time I looked there was no application of ANNs which couldn't be solved more efficiently by other algorithms ... and the best ANNs used spiking neurons with Hebbian learning which are not amenable to efficient digital implementation.

Is it possible that last time you checked was a long time ago? Deep neural networks are again all the rage now (i.e. huge teams working with them at Facebook and Google) because

  1. (1) They have resulted in a significant performance improvement over previously state-of-the-art algorithms in many application tasks,
  2. (2) Although they are computation-heavy, they are amenable to massive parallelization (modern computational power is probably the main reason why they have improved singificantly with respect to ANNs of the 80-90s, given that the main architecture itself has not changed a lot, except possibly for the "convolution" trick which effectively introduces hard-coded localization and spatial invariance).

To be fair, it always seems to me that (1) and (2) are very closely related. CNNs that won recent computer vision benchmarks are the only methods that used so much processing power so far. Not that they're less efficient than other, tough. It's just that I would love to see other methods with that many engineering, tunning, dedicated computational power and how they compare.
Also, not that when it comes to classification, the standard is to throw the last layer and train a linear SVM on the penultimate layer, which also show that CNNs alone are not enough.

Comment: Predictable (Score 2) 441

by lorinc (#48825725) Attached to: Why We Have To Kiss Off Big Carbon Now

In the long run, it will fade away because most of the grouwth has already been consumed. That being said, trade is chaotic in nature, and short term prediction is difficult ("especially when it's about the future"), but in the long run, the trend is well known.

Sometimes, I like to think that the "Limits to growth" report will be regarded in some distant future as our epoch's Eratosthenes calculations.

Comment: Re:Well he would say that. (Score 0) 894

by lorinc (#48819847) Attached to: Pope Francis: There Are Limits To Freedom of Expression

P.s We don't get our morals from religion, my observation is that quite often "religious" people have less ethics and morality than atheists.

Well, to be completely honest, people with an imaginary friend are usually send to psychiatric hospitals. Unless there are several millions/billions of them having the same imaginary friend...

Comment: scientific stars wannabe (Score 2) 227

by lorinc (#48806179) Attached to: Lawrence Krauss On Scientists As Celebrities: Good For Science?

The problem, is that scientific research is now like music was in the 80s. People are much more interesting in writing the article that will be cited 1k times, like people were looking to write that single getting sold 1M times, than actually improving common knowledge.

Well at least in computer vision, I do have this impression.

Comment: Re:It's hard to take this article seriously (Score 1) 628

by lorinc (#48642617) Attached to: What Happens To Society When Robots Replace Workers?

What if it is possible to be completely autonomous with machines? Machines that take care of your food, clothes, transportation, anything you ever need or want, but machines that require a massive amount of wealth to acquire first. What if the future of automation is the mere 1% that live in a libertarian utopia (kind of The dancers at the end of time by M. Moorcock) and the remaining 99% struggling to survive? what if the over-concentration of wealth you are observing right now is just the first step towards that kind of future?

Isn't it plausible that over-concentration of wealth is a natural consequence of automation? To me it seems intuitive that the concentration of wealth into singular points is the end goal of automation.

Comment: The New Magic (Score 3, Insightful) 74

by lorinc (#48396275) Attached to: Machine Learning Used To Predict Military Suicides

Stop speaking of machine learning as if it's a new kind of black magic. I know it sounds better than "using a mathematical algorithm" or "performed statistical analysis", but to me it sounds as ridiculous as the "quantum whatever" of the 90's. Seriously, ML is being hyped beyond reasonable.

Comment: Re:Home storage (Score 1) 488

by lorinc (#48366639) Attached to: Denmark Faces a Tricky Transition To 100 Percent Renewable Energy

This is nothing compared to the price of the house. Let the company that builds the house install them, like they install isolating windows (which are more expensive than that, btw). Or rent it from your electricity provider. Yeah it's expensive for an egoistic individual, but it is not for the entire society.

Comment: Re:I disagree. (Score 1) 145

by lorinc (#48213585) Attached to: Machine Learning Expert Michael Jordan On the Delusions of Big Data

You should probably take some lectures in computer vision, it would change your view on it. It's either that, or you have a misconception about what a human does when he's learning.

I'll take the Turing view on humans: big and horribly complex machine running a big and horribly complex algorithm. A part of this algorithm and its dedicated hardware is something we call "vision". Of course, it's a big an clunky part, and we even don't know its exact boundaries.

Now suppose you have a computer that does run an algorithm which is equivalent to the one of the human. Wouldn't you call it "computer vision"? I think that's a pretty good name that reflects correctly what the thing does.

Here comes the tricky part: we don't have the algorithm yet, nor do we know all the things it should do. Hell, we don't even now how to measure the equivalence with the human one!

But things are advancing and we're getting more and more pieces that seem to behave closer to the human algorithm (same inputs, same outputs). Some pieces are easy (depth estimation from calibrated stereo cameras), some are more involved but we can do it pretty well (OCR), some are really hard and we begin to do them not so badly (object recognition) and some are just so crazy nobody is even working on it (animal identification comes to my mind. Example: which of the wolves in the pack is Titus?).

I have no doubt we'll have one day a collection of algorithm that can do anything a human can do with his vision, whatever the environment. If that's close or far away is still debate though. I guess it will be much quicker than anyone expected, myself included. Will that collection be the same as the one in the human brain, i.e., will we have deciphered the human algorithm? No. That's not even a question in computer vision, that's a question of cognitive sciences. Although some parts can be inspired by what cognitive science tells us about the human mind (see, e.g., the recent developments in deep neural nets).

Suppose you now have that algorithm, wouldn't you sell it as "a complete solution comparable to a human" to quote your words? Of course you would, that's exactly what you tried to achieve from the beginning. Now, if a magazine sells it a already done while it still is being research, blame the journalist, not the scientists. It's been 50 years that journalists have sold nuclear fusion as our energy revolution coming next year, and still you wouldn't say physicists in the field are only good at marketing without strong stuff behind them.

(disclaimer: I also work in the field)

Comment: Re:Limited Vision (Score 1) 287

by lorinc (#48210569) Attached to: Will the Google Car Turn Out To Be the Apple Newton of Automobiles?

I don't agree with you. Scientific expansion in ML is of exponential growth: what took 20 years to achieve will take only 10 to be doubly improved. When I see the state of the art in computer vision 2014, I have almost no doubt the vision problems associated with automatic cars will be solve in a fewer amount of time that anyone expected.

The Tao is like a glob pattern: used but never used up. It is like the extern void: filled with infinite possibilities.