Beta

Slashdot: News for Nerds

×

Welcome to the Slashdot Beta site -- learn more here. Use the link in the footer or click here to return to the Classic version of Slashdot.

Thank you!

Before you choose to head back to the Classic look of the site, we'd appreciate it if you share your thoughts on the Beta; your feedback is what drives our ongoing development.

Beta is different and we value you taking the time to try it out. Please take a look at the changes we've made in Beta and  learn more about it. Thanks for reading, and for making the site better!

MIT & Harvard On Brain-Inspired A.I. Vision

Soulskill posted more than 4 years ago | from the i-can-see-clearly-now dept.

Software 27

An anonymous reader writes with this excerpt from TGDaily: "Researchers from Harvard and MIT have demonstrated a way to build better artificial visual systems with the help of low-cost, high-performance gaming hardware. [A video describing their research is available.] 'Reverse engineering a biological visual system — a system with hundreds of millions of processing units — and building an artificial system that works the same way is a daunting task,' says David Cox, Principal Investigator of the Visual Neuroscience Group at the Rowland Institute at Harvard. 'It is not enough to simply assemble together a huge amount of computing power. We have to figure out how to put all the parts together so that they can do what our brains can do.' The team drew inspiration from screening techniques in molecular biology, where a multitude of candidate organisms or compounds are screened in parallel to find those that have a particular property of interest. Rather than building a single model and seeing how well it could recognize visual objects, the team constructed thousands of candidate models, and screened for those that performed best on an object recognition task. The resulting models outperformed a crop of state-of-the-art computer vision systems across a range of test sets, more accurately identifying a range of objects on random natural backgrounds with variation in position, scale, and rotation. Using ordinary CPUs, the effort would have required either years or millions of dollars of computing hardware. Instead, by harnessing modern graphics hardware, the analysis was done in just one week, and at a small fraction of the cost."

cancel ×

27 comments

FP (-1, Troll)

Anonymous Coward | more than 4 years ago | (#30335134)

It's a celebration, bitches!

Re:FP (-1, Flamebait)

Anonymous Coward | more than 4 years ago | (#30335308)

I fucked your dead great grandmother!

Re:FP (-1, Flamebait)

Anonymous Coward | more than 4 years ago | (#30335594)

Yeah, well that means that you got fucked worse, 'cause she liked it!

Inconsiderate (3, Funny)

Narpak (961733) | more than 4 years ago | (#30335314)

Instead, by harnessing modern graphics hardware, the analysis was done in just one week, and at a small fraction of the cost.

How inconsiderate. Think about all the potential engineers, administrators, janitors and etc, that would have been needed to do all that work the slow way; thus creating jobs for many for years to come. With one swoop all that potential future effort was made redundant, once again "researchers" have proven that they are unable to see the big picture!

Real soon now (0)

Zerth (26112) | more than 4 years ago | (#30335328)

One of these guys will read about GA, realize this has been done before in other problem spaces, and already has a name.

Try not to get stuck on a local maxima!

Not Really a GA (1)

rayharris (1571543) | more than 4 years ago | (#30336536)

They didn't really use a GA. They had a genome that described the structure of the neural net they wanted to test, but they didn't "evolve" the population through any process of mutation or crossover. They just kept generating new random individuals until they had a good one.

It's like only doing the first step of a GA, but you keep generating random starting points until you find one who's fitness is fairly high (although they did a uniform sampling over all parameter values for their starting point, not quite completely random).

From the paper, it took 23 PlayStation 3's one week to generate, train, and test a population of 7500 individuals.

Even so, they still beat the best hand-designed solutions. Imagine if they had implemented a true GA and just let that system keep on running.

Oh, and they used Python! I'm encouraged. I'm about half-way through translating the C# implementation of the HyperNEAT algorithm into Python. Next, I'll have to get some PS3s and implement a distributed PyCuda HyperNEAT system.

Re:Not Really a GA (2, Informative)

linhares (1241614) | more than 4 years ago | (#30337094)

Ok, just skimmed the paper [ploscompbiol.org] . First impressions: it's a good idea. The problem I see is that, after finding a great model, they have absolutely no clue as to why that one works. That is, a functional theory isn't improved by this kind of work; though it is indeed promising and the theory can be improved if someone can understand what the heck that model is doing.

Monte Carlo (0)

Anonymous Coward | more than 4 years ago | (#30340750)

They did not use a genetic algorithm, instead they took the Monte Carlo approach. Not a bad approach if you don't mind making the results better by hand, but not as good (or the same) as a GA. Also not a new approach.

Yeah... (-1, Offtopic)

Anonymous Coward | more than 4 years ago | (#30335540)

...but when can I have fricken 'robot' sharks with fricken laser beams coming out of their foreheads?

GPU processing was news 2 years ago (0)

Anonymous Coward | more than 4 years ago | (#30335674)

Using a GPU to do heavy processing isn't really revolutionary anymore- it's been done before, in many different applications. It bothers me how everytime new research comes out that uses this technique, journalists sensationalize the GPU aspect of it, often taking away from the actual breakthrough.

Is this the result of researchers themselves going on about how CUDA made their lives easier, or the journalists saying 'woah woah, you did this with VIDEO GAME stuff? Tell me more!'

Low hardware (2, Interesting)

gmuslera (3436) | more than 4 years ago | (#30335756)

Eyes, brain raw power, could be considered somewhat "low" technology, But you need to be smart to implement a pattern recognition engine (and integration with existing data) as the brain have. Think that you can have "vision" with something far less precise than eyes (with i.e. this [seeingwithsound.com] and similar low res devices).

How much power requires that pattern recognition? By standards approachs probably a lot, but the approach they seem to use there (like in compare how much fits what they have with thousands of candidate models) could require less, and far better if you use for that hardware that are more adequated for that task.

Re:Low hardware (1)

rayharris (1571543) | more than 4 years ago | (#30336664)

What I think we're going to find is that one type of system won't be sufficient for us create an AI. Early work was done on symbolic systems. Those eventually worked pretty well on idealized domains. They fell apart when they tried to interact with the real world. Neural nets are starting to handle simple real worlds tasks, but can't handle complex domains.

My thought is that we'll see something like this:

Audio/Visual/Etc. Input --> Neural Net-based symbol extractor --> Symbolic Planning and Decision System --> Neural Net-based motor control --> Motors

In this integrated system:

1. Neural nets will be developed to process raw input and generate a set of symbols.

2. Expert systems, symbolic planning systems, knowledge bases, etc. will work together to reason about the input and make a decision what action to take.

3. Another set of neural networks will then translate actions in a symbolic form into control inputs for servos and actuators.

I have no idea how the Asimo robot works under the hood, but it seems to be something similar to this ( http://world.honda.com/ASIMO/technology/intelligence.html [honda.com] ). Anyone have any more details about how Honda implemented the different recognition capabilities?

Genetic algorithms? (2, Insightful)

jjh37997 (456473) | more than 4 years ago | (#30336460)

The team drew inspiration from screening techniques in molecular biology, where a multitude of candidate organisms or compounds are screened in parallel to find those that have a particular property of interest. Rather than building a single model and seeing how well it could recognize visual objects, the team constructed thousands of candidate models, and screened for those that performed best on an object recognition task.

Without reading the article, because that would be silly, this sounds a lot like using genetic algorithms. Not actually a new technique.

Re:Genetic algorithms? (0)

Anonymous Coward | more than 4 years ago | (#30336554)

Correct. It is also frustrating that (I just skimmed it though) they don't actually say what the best algorithm is!

It is cool they summed their best 5 and got an even better one (because they are not very similar to each other). And it works on different kinds of data (real photographs) while the state of the art until now had to be trained for a problem space.

Nice way to get free hardware from Nvidia!

Re:Genetic algorithms? (0)

Anonymous Coward | more than 4 years ago | (#30336572)

If anything with the word "organism" in it sounds a lot like "genetic algorithms", then yes. In GA, the team wouldn't really be 'screening' anything.

Not really a GA (1)

rayharris (1571543) | more than 4 years ago | (#30336696)

As I posted in a reply above, they didn't really use a GA. There were no mutations or crossover. They just kept generating new random individuals until they had a good one.

Re:Not really a GA (0)

Anonymous Coward | more than 4 years ago | (#30337414)

It makes you wonder why they didn't actually use mutations, crossover and selection. Random searching of such a vast space is incredibly inefficient. They would have to have some form of selection to weed out potential candidates and breed them -- and if they didn't, they should have.

Still not a new idea (1)

raftpeople (844215) | more than 4 years ago | (#30337448)

I personally have been doing this as well as the same thing with mutations for 6 years in an artificial life/neural net simulation. And I'm just a hobbyist (many researchers have and are doing all kinds of this type of stuff). It's definitely a powerful technique and fun to read about their success, but hardly new.

Re:Genetic algorithms? (1)

RandCraw (1047302) | more than 4 years ago | (#30341560)

First of all, the authors have chosen the wrong metaphor. HTS is about sending a large number of different organisms through the same assay, looking for candidate targets that produce a positive response to that one test. The folks at Rowland are doing the inverse. They're sending a large number of assays at a single target (many candidate algorithm variations (assays) seeking to recognize an image or a specific pattern therein).

Second, they're generating candidate algorithms, not just algorithm parameters. That's more like genetic programming than genetic algorithms.

Third, the kind of algorithms they describe sound more like neural nets with a large combinatoric variation of units, layers, feedback components, and sundry other possible variants. These seem to me to be both a biologically plausible mechanism for a biologically inspired vision model, not to mention a viable candidate for the pattern recognition of a wide range of vision targets. GA-based mechanism would be too symbolic and boolean to be biologically plausible, and too rigid in matching specific input patterns.

Fourth, a better metaphor for the parameter sweep they describe in the video is monte carlo methods (generating and testing many many parameter combinations using a supercomputer), not HTS, HCS, GP, or GAs.

And no, I haven't read their research papers yet. But I definitely will. Interesting stuff. I wonder if they have/will tap-ped into David Marr's bio-inspired vision models & theories?

Re:Genetic algorithms? (1)

acheron12 (1268924) | more than 4 years ago | (#30342000)

Third, the kind of algorithms they describe sound more like neural nets with a large combinatoric variation of units, layers, feedback components, and sundry other possible variants. These seem to me to be both a biologically plausible mechanism for a biologically inspired vision model, not to mention a viable candidate for the pattern recognition of a wide range of vision targets. GA-based mechanism would be too symbolic and boolean to be biologically plausible, and too rigid in matching specific input patterns.

Actually, using GAs (an optimization method) to train a neural network (an optimization problem) is a fairly common technique [google.com] .

But does it work? (1, Insightful)

Anonymous Coward | more than 4 years ago | (#30336888)

It seems to me that they are just using random functions to see what works best. But what they neglect to say is how good their best functions do. What percentage of of identification is correct. The assumption is that the brain uses some mathematical function for its processing which may not be the case.

Re:But does it work? (1)

Neuroelectronic (643221) | more than 4 years ago | (#30337370)

Do you have a link to the original PLoS article? I could not find it.

Re:But does it work? (0)

Anonymous Coward | more than 4 years ago | (#30339574)

The answer is underwater crayfish and their eyes. Or the humble fruitfly.
They have small brains. So color and spectrum are important.
Much cheaper to use overlays and do a binary chop of image candidates.

The military is trying to make such eyes/sensors, but so far insects and sea life are smarter - and refuse to give up their secrets.

Combinatorial Chemistry (1)

Teufelsmuhle (849105) | more than 4 years ago | (#30340652)

Combinatorial chemistry [wikipedia.org] techniques seem to be the inspiration here. Also known as trial and error (albeit in a rapid well-organized fashion). Not exactly a new idea, but this is an interesting new implementation.

echolocation (0)

Anonymous Coward | more than 4 years ago | (#30340822)

First thing that comes to mind is echolocation when I think of ways to measure distance. Visualy it would take much more to process and recognize shapes if I take an educated guess and simulate the process. It would have to deal with various lighting conditions, and geometries. Sending a ping signal like a submarine can be measured equally as well, if not better than I expect a visual system to do it. In fact our eyes work in rather the same way, collecting light that bounces off objects.

huh (1)

Entropy997 (1694668) | more than 4 years ago | (#30365800)

Who knows if Nvidia OR AMD (think: Fusion) has been funding this research?
That part about magic cost reduction sounds a little funny.
Also, I think it should be said that a decent single core graphics card might cost $150 vs. one of those power-wise full featured $300 PS3's with the Cell processor (not the mention a stream-processor carrying Nvidia/ATI video card).
Check for New Comments
Slashdot Account

Need an Account?

Forgot your password?

Don't worry, we never post anything without your permission.

Submission Text Formatting Tips

We support a small subset of HTML, namely these tags:

  • b
  • i
  • p
  • br
  • a
  • ol
  • ul
  • li
  • dl
  • dt
  • dd
  • em
  • strong
  • tt
  • blockquote
  • div
  • quote
  • ecode

"ecode" can be used for code snippets, for example:

<ecode>    while(1) { do_something(); } </ecode>
Create a Slashdot Account

Loading...