Slashdot Mirror
GNU Project Introduces Gneural Network AI Package (gnu.org)
jones_supa writes: The GNU free software project is introducing a new neural network computation package called Gneural Network. The GNU project has been impressed by the work of Google, IBM, AlphaGo and Watson on the field of artificial intelligence. However, the GNU project sees that the fact that only companies and labs have access to this technology can represent a threat: "First of all, we cannot know how money driven companies are going to use this novel technology. Second, this monopoly slows down Progress and Technology." This is why the author, Jean Michel Sellier, decided to create Gneural Network and release it under the GNU GPL license. In the current release (version number humbly set to 0.0.1), it is a very simple feedforward network which can learn very simple tasks such as curve fitting, but the development team plans to deliver more advanced features very soon. They are already spending efforts to implement a network of LSTM (long short term memory) neurons for recurrent networks and deep learning. Learning reinforcement techniques are also planned.
TensorFlow is already opensource and does more...
"However, the GNU project sees that the fact that only companies and labs have access to this technology can represent a threat"
That is not a fact at all. There are tons of open source neural network libraries and tools and even tons of open source neural network libraries that provide recurrent network and deep learning features. Just a 30 second search gives me this list:
http://deeplearning.net/softwa...
"a very simple feedforward network which can learn very simple tasks such as curve fitting"
This is NN101 stuff and I'm sure hundreds if not thousands of college students have made something similar.
Nothing to see here. Move along.
Not to shit all over their good work, but the most successful projects under the aegis of GNU, succeed _despite_ the neckbeards in Boston, not because of them.
If it's anything like the utter embarrassment of HURD (w.r.t. Linux kernel), these guys will still have some slow piece of crap that barely compiles, a long time after something else has built something far better, with a freer license, at a point that deep-learning is baked into absolutely everything around us -- just like Linux.
I think GNU themselves are too slow, dumb and doctrinaire to ever produce anything of value or impact ever again.
The only major difference between BSD and GPL licenses is that BSD allows open software to be closed, so really you're arguing in favour of closed software. Whatever are you doing on Slashdot?
But the illogic of your position runs deeper still. The whole point of TFA and of Gneural is to provide an open neural net because closed ones are already plentiful , so the only perceived "benefit" of BSD (using the term loosely) is precisely what Gneural is trying to balance. This makes your desire for BSD licensing so that even more proprietary software can be made totally miss the point of the project.
The idea that "the fact that only companies and labs have access to this technology can represent a threat" is patently absurd. Theano, Caffe and Torch are all open source and even Google has open-sourced its Tensor Flow platform which makes it easy to build new tools and run then, fast, on all the GPUs you can find. If you need to do this at scale and you're not the size of Google or IBM you can use Amazon's Machine Learning for AWS. There are many, many higher level toolkits out there that are available under licenses that are much less restrictive than GPLv3.
If intelligent life is too complex to evolve on its own, who designed God?
Does anyone know how this is different from the other open source neural nets that exist?There have been tons of these over the last 40 years. AFAIK most of the algorithms originate in academia and stay there where open sourcing is the norm.
The GNU project should do a bit more background research before starting new projects. Here are some links to open source deep learning tools. These are the same tools and libraries used by those "money driven companies" in their projects, including AlphaGo:
Caffe, widely used C++ deep learning framework.
Theano, widely used Python deep learning framework.
Torch, the software used by Google, AlphaGo and Facebook.
TensorFlow, Google's large scale machine learning framework.
CNTK, Microsoft's deep learning toolkit.
I miss the good old days where companies and people announced interesting things they've done rather than interesting things they're thinking of doing.
For someone reason I read the article and looked at the code. The project is under CVS. There is no neural network interface, only a main method with supporting code that creates one hard-coded NN. The serialization format of the NN is a custom, non-standard format which you have to manually scan and parse characters to read back in. Why write custom parses when you could more easily use xml, json, csv, ini, etc...? I'm mainly a Java and Lua guy, but I do have some C++ experience. In C, is it common to put all the function code in the header files? Because that's what's in there. Part of the documentation refers to his mother (and to be fair, an equal amount to his dad too) and the rest of the documentation says to read the code. There's no plan or suggested roadmap for how this library is going to turn out. How can people contribute if they don't know what it's supposed to look like?
I think I'll forget about this project forever and use one of the much higher quality FOSS neural network libraries, such as the ones in Weka or the 13 year old FANN library. Some advice, just because you're not constantly hearing people gloating over their awesome bits of stuff doesn't mean it doesn't already exist. There's over 7 billion people currently alive in the world right now. Chances are someone beat you to it.
Good luck. You need it.
Yeah, but you know, it is the Apache 2.0 opens source license. It is not good enough. Every developper shoud have its name on every newspaper headline for releasing its implementation of the first exercise of its machine learning course under the GPL. He want to put its library against actual behemoth who can do way more and are for some even GPU accelerated. I am sorry, but if the GNU project should promote such library only when it could be compared to other such libraries such as TensorFlow or any other currently available library. I really like the GNU project work, but on this one, they really hurt my feelings about them.
And Cafe, Theano and Torch 7 are open source and do even more.
There are a bunch of really good implementations, and several of them have nothing to do with companies, but do have an existing community developing them. This GNUish project doesn't seem to have any advantages and is arriving late to the game.
The GPL is about the freedom of the user, not the developer. It is designed to ensure that users always have access to the source of software they run and any updates. There is no such guarantee from BSD or any other licence.
1990 called and wants its version control system back. I'd go poking around in their version control to at least determine the implementation language, but... nah.
I'm trying to teach myself to set people on fire with my mind... Is it hot in here?
That's Caffe. You won't get very far searching on Cafe, unless you're looking for a coffee shop. :)
The GNU project has been impressed
Yeah like that's difficult, they were impressed by my footpath of soil and earthworms; they were impressed when I replaced my home with an off-grid refrigerator box; they were impressed when I replaced my computer with a grid of raspberry pis linked to recycled cell phone screens linked together as pixels in a giant grid array plastered to the wall of my refrigerator box; they were impressed when I stopped using deodorant; they were impressed when I convinced all my friends that Microsoft was the spawn of satan on earth.
But then I got a job, friends, and a family, and they acted like they didn't know me anymore.
a whole bunch of disconnected projects with duplicated effort and little actual result is proving the point that a unified project will achieve better results
So soon I will be able to use this to help make my ultra-drone army even more effective at killing all the humans! Now I just need to perfect my human glucose-based power harvesting, and my biological harvested bioprinter!
I couldn't find the source code repository. It might just be because I am clueless on how to find them for GNU projects. Anyone devise the subversion or git or darcs or whatever it is location?
"Recursive bipartite matching"- try it!
It would be lovely to imagine that you're right. Unfortunately, the opposite has been true for at least a decade.
RMS won the code openness wars in the name of user freedoms, but now the corporations are taking back the land that was won, and they're doing so in the name of profit --- their own exclusive profit. In no case whatsoever do they allow communities or other companies to alter the course that that they have set for their own BSD or MIT-licensed software. Their code is not open in the sense that community software is open. It is not allowed to evolve to reflect user-oriented needs, but is very tightly controlled instead to meet their business goals alone. This is not the Bazaar, but the Cathedral.
Even the concepts of interoperation and federation that have been a sort of "Prime Directive" of the IETF Mission Statement have now been dismantled by the corporations, every one of them intent on making their own walled garden instead of defining federated services that interoperate freely. Openness is under attack on numerous fronts simultaneously, all in the name of profit.
BSD and MIT licenses are being used to deny community freedom over software instead of to extend it. RMS's idealism is needed now more than ever to balance the power of the corporations, who are most definitely not on our side.
Full text of the BSD 3-clause license:
(The BSD 2-clause license is almost identical, just lacking the 3rd clause concerning "endorsement".)
As should be obvious from the above, these licenses impose no condition whatsoever that the licensed code must remain open. They even go the extra mile and state the condition that must be met if the code is closed:
Hopefully that is clear --- the source code can be closed off completely (not a single program statement need remain open), as long as the text just mentioned is supplied in accompanying materials.
GNEW! Chewy GNoUgat Beowulf Cluster Candies! The first Free and Open Source Candy! Fully user customizable!
Tired of proprietary H-Bombs? Why not build GNUclear weapons! The first Free and Open Source Weapons of Mass Destruction!
Make your very own life! With F/LOSS DeoxyriboGNUcleaic Acid strands! Go on, PLAY GOD a little!
Want something to listen to, that's groovy but free? Try GNUwave music!
Etc.
I guess we should count oursevles lucky that they chose a gnu for their mascot, and not an ASS. Otherwise everything would start with that. Hey, what software are you using to manage the movement of electrons through your semiconducting material? Why, ASSHoles of course. Have you tried the new version of ASS/Linux? It's made to ease the transition from using Windows by having a very similar interface, from ASSHAT Software. They also provide support! There's a new port of a popular arcade game, have you tried it? It's called ASSteroids...
Be happy they chose a gnu.
No. The GPL is about freedom of the derivatives of the software. For the current incarnation/generation/instance of the software BSD is freer. The GPL guarantees that the derivative software will also be free.
Sometimes one is a better choice than the other, but neither is uniformly preferable.
I think we've pushed this "anyone can grow up to be president" thing too far.
This is why the author, Jean Michel Sellier, decided to create Gneural Network and release it under the GNU GPL license.
Let's be honest. It was purely for the pun.
Check for yourself. Straight from the horse's (or gnu's, if you so prefer) mouth: http://www.gnu.org/licenses/li.... See there Apache, Artistic 2.0, several BSDs, CECiLL, and lots of others you never heard about listed under free.
Please, research your stuff before spouting out.
Machine learning is software generated by statistical algorithms fit to lots of data. Without the training data, the algorithms alone are quite useless. Pre-trained networks are essentially closed source, because the source is the training data.
There's lots of open source code for this work already. It boils down to who has access to the data. Tesla can turn on autopilot to collect data from its entire fleet for millions of miles traveled. Google doesn't have a fleet, so it wants to collect so much data with Android Auto, automakers are walking away.
GNU, well, they've got some algorithms, just like everyone else.