GNU Project Introduces Gneural Network AI Package (gnu.org) 95
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.
Re: (Score:2)
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
Re: (Score:3)
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.
Re: (Score:2)
You're not making sense (Score:4, Insightful)
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.
Re:You're not making sense (Score:5, Insightful)
So paraphrasing:
BSD license = you are free to do with the software what you want.
GPL license = we are going to dictate what you may do and what you may not, which products it is acceptable to build, etc.
Sounds like one of those is much freer than the other...
BSD is freer in the first generation sense but when someone takes BSD and expands on it (for example MacOS), other people are then cut off from those added improvement. It would be like a free public library where no one was required to return the books. Yes, the books are more free for the first person that checks them out but if the first person puts them on their shelf at home, those books are now a lot less free for future users. In this case, they are obviously worried that as machine learning becomes more and more important they don't want the market cornered by a handful of commercial companies negating the last 20 years of their progress. If instead of paying a "microsoft tax" to use windows everyone has to start paying a "google tax" for the machine learning to make their computer usable then we are back where we have started from.
Re:You're not making sense (Score:4, Insightful)
Except that the original project still exists. The BSD license is like an electronic library where you can check out the books for free. You don't get access to my margin notes unless I choose to return my copy of the book, but the original is still there and you can read it and make your own notes as you please.
Re: (Score:3)
The original project is not what matters. What matters is the version that is in use today. You've been around long enough to know this. Having a fully documented set of instruction to build a carriages doesn't make your car's OS any more open source.
Re: (Score:2)
You say that as if that opinion is the only one that can exist...
Re: (Score:2)
I say that presenting the BSD argument without countering the way known retort is essentially dishonest. The poster knew the retort and is obligated to present it when discussing the opposing views.
Re: (Score:2)
If you're using the book model, it's more like you're selling your annotated version without bothering to pay royalties back to the original author.
A lot of GNU authors were commercial or BSDists who were sick of seeing their work effectively pirated (either literally, or by having companies fold their BSD work in and sell it for profit). At least with GPL there's an obligation to pass an annotated copy back to the library if you start selling it and some teeth which make it painful enough for companies whi
Re: (Score:1)
then cut off from those added improvement
Which WERE NEVER BSD Licensed, never public, and your entire argument falls apart right there at that statement. You can't take away something you never had. They didn't 'close' something that was 'open' they mixed 'open' and 'closed' together, and you're just pissed off you can't get it for free when you, personally have made absolutely 0 contribution to it.
It would be like a free public library where no one was required to return the books.
No, Sony/BMG/MPAA (who ever you work for, they spew that shit), its not like that. One someone takes a copy of the source, the original is still the
Re: (Score:2)
So which version of BSD can run Office 2016 for Mac? None. That's the point. Even though almost all of Darwin is open source that doesn't change the fact that OSX is closed source.
Re: (Score:2)
BSD license = you are free to do with the software what you want.
GPL license = we are going to dictate what you may do and what you may not, which products it is acceptable to build, etc.
Sounds like one of those is much freer than the other...
BSD License = you are free to enslave your users.
GPL License = you are not free to enslave your users.
As a user I know which seems freer to me.
Re: (Score:2)
BSD license = you are free to do with the software what you want.
Depends on what BSD license you're talking about. The original BSD license required an advertising clause.
Re:You're not making sense (Score:5, Insightful)
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?
Implying one have to be a GPL fanatic to see any value of /.? That's of course bullshit bordering on trolling.
But you are also wrong about the BSD licence, it doesn't allow open software to be closed - it allows open software to be _forked_. That property it shares with the GPL where the copyright owners can fork their code to be licensed however they want. So the difference is who can fork the code, all (BSD) or the copyright owners (GPL). So which is more free?
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.
And you plainly doesn't understand the BSD licence thus missing the point of the post.
Re:You're not making sense (Score:4, Interesting)
No, he is arguing in favor of freedom. Freedom is meaningless if it doesn't include the freedom to do things others disapprove of. Software licenses are a mechanism for achieving freedom. RMS believes that the best way to achieve freedom is through restrictive licenses backed by the coercive power of the state (copyright, courts, police). Other people believe that the best way to achieve freedom is through non-restrictive licenses and by example. Which of those mechanisms is better is an empirical question, but at this point, it looks like RMS is losing the debate.
Or, alternatively, you might also say that RMS has won the debate but just doesn't want to admit it: the GPL was never intended to be a permanent solution, it was intended to be transitional, until people have learned the benefits of sharing and do it freely. Well, that's what's been happening, with the vast amounts of software released under "do anything, just don't sue us" licenses by private companies.
Personally, I thank RMS for his original contributions and getting people use to the idea of sharing software, but I think his task is done, and for now, at least, the BSD and Apache licenses are better choices for most projects.
Too generous an interpretation (Score:1)
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
Re: (Score:2)
You're arguing in favor of the freedom of people. Freedom is fleeting if they can slip chains on you at a latter date. Software licenses are a mechanism for achieving freedom. The GPL keeps the codebase free by protecting it from those who would slip chains on it. Other people prefer to be free to bind your works with even more perilous licenses. Hey, it worked out pretty well for BSD and Apple's OS X. Through the goodness of Apple's heart and bottom line they made contributions back from OS X to BSD. But a
Re: (Score:2)
Someone releasing a closed source version of an open source project doesn't affect the freedom of the open source project.
Linux is an old project, as is gcc. If you look at newer open source projects, they are (1) largely started by private companies and (2) increas
Re: (Score:1)
You can't 'close' BSD licensed source code, if you think you can, you're a moron and have no fucking clue how software licenses work at all.
You can make changes to it and close the changes you made, but you can not take the original code and close it.
So basically, you are flat out lying.
That raise the question are you stupid or intentionally spewing ignorance misinformation?
Re: (Score:2)
You can make changes to it and close the changes you made, but you can not take the original code and close it.
That's probably what he means.
So basically, you are flat out lying.
No, just a communication error.
Re: (Score:3)
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.
Re: (Score:2)
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.
An open source NN library. How novel! (Score:5, Informative)
"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... [deeplearning.net]
"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.
Re:An open source NN library. How novel! (Score:4, Insightful)
Yep, nothing novel at all, even in conception. The real trick with deep learning software is figuring out how to integrate them into knowledge bases and provide useful training and feedback mechanisms. Honestly, creating the neural network software is the easy part, because there's a ton of academic research that tells you *exactly* how to do it. I actually downloaded and examined the code, and while it looks reasonably clean and functional, we're not exactly talking about a huge amount of work to replicate it - it's just a few hundred lines in total.
Moreover, when any software package starts with a section on "The Ethical Motivations" for its existence, it strikes me as the wrong sort of motivation altogether. The real motivation should be "I want to solve some interesting problems", and THAT will drive the design. This sounds like a pretty typical academic exercise, and as such, probably is not going to amount to much, other than as a starting point for some student projects here or there. But even that is dicey, as naturally, there's no documentation at all - just a readme file telling you to look at the source code to figure out how it works. Odds are pretty good that documentation is never actually written for it, because that's a hell of a lot less fun than writing the code.
Sorry, I really do love this sort of stuff, but it's a little hard to get excited about the project when its exactly the same sort of code I was tinkering with as an undergrad student decades ago. That anyone is actually comparing a few hundred lines of relatively simplistic C code to IBM or Google's machine learning projects is disingenuous at best, borderline insulting at worst.
Re: (Score:2)
I'm curious. Is the ethical motivations preamble longer or shorter than the actual code?
Re:An open source NN library. How novel! (Score:5, Funny)
Let's just say that the ethics preamble (at five lines) is longer than the library's current documentation.
Re: (Score:2)
Well, they do need fresh blood to finish off GNU/Hurd, this is one way to get it...
Re: (Score:2)
Well, they do need fresh blood to finish off GNU/Hurd, this is one way to get it...
I don't believe that GNU/Hurd was ever intended to be finished. I think the idea was to have a permanent Work In Progress.
Re: (Score:2)
No, you've misunderstood. To get Hurd working they literally need blood: like actual blood. Without signing the soul in the right way you cannot bind it into the software.
Re:Vapour? (Score:5, Interesting)
If it's anything like the utter embarrassment of HURD (w.r.t. Linux kernel)
The HURD ceased to be a prority GNU project decades ago. The Linux kernel meets the FSF requirements (GPL) and so completes teh GNU operating system. What's left the the research project of a few people who want to see if they can make a super-unix with a microkernel.
That is not in any way shape or form embarressing.
I think GNU themselves are too slow, dumb and doctrinaire to ever produce anything of value or impact ever again.
I, for one want to see concepts in C++ as soon as possible. Gue'ss who's leading the charge on that one?
Re: (Score:2)
That is not in any way shape or form embarressing.
Even when you consider that several projects have already succeeded in replacing the Linux kernel with existing, proven microkernels?
Re: (Score:2)
Even when you consider that several projects have already succeeded in replacing the Linux kernel with existing, proven microkernels?
Genuine question - citation please!
Re: (Score:2)
Of the top of my head K42 and a few L4 projects (Wombat/Iguana) does that. Don't know if calling it "replacing" is correct though.
Re: (Score:2)
Even when you consider that several projects have already succeeded in replacing the Linux kernel with existing, proven microkernels?
Nobody has replaced Linux with a microkernel. Some people have run Linux as a service inside a microkernel, though.
Re: (Score:2)
Even when you consider that several projects have already succeeded in replacing the Linux kernel with existing, proven microkernels?
Right, so we're seeing those microkernels running all the world's supercomputers, a good fraction of the world's servers, a lot of routers and the majority of smartphones? Nope? Nope.
Nothing has replaced Linux.
Re: (Score:2)
What's left the the research project of a few people who want to see if they can make a super-unix with a microkernel.
Most people have realized that the answer is "no".
Re:Vapour? (Score:4, Insightful)
NN have been around for 40 years. Lots of people have built stuff already.
Mainly FSF is a political organization not a software shop. They did a lot of good work, and they failed on some projects. Lots of top quality people couldn't keep up with the Linux kernel no embarrassment in that. The person who gets the bronze in the olympics is not a failure.
"Ethical Motivations" (Score:5, Informative)
The idea that "the fact that only companies and labs have access to this technology can represent a threat" is patently absurd. Theano [deeplearning.net], Caffe [berkeleyvision.org] and Torch [torch.ch] are all open source and even Google has open-sourced its Tensor Flow [tensorflow.org] 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 [lorienpratt.com]. There are many, many higher level toolkits out there that are available under licenses that are much less restrictive than GPLv3.
Re: (Score:2)
There are many, many higher level toolkits out there that are available under licenses that are much less restrictive than GPLv3.
Aaaand ... you found the problem. See, wasn't that hard, was it?
The GNU people don't recognize anything as 'free' if it's not licensed under GPL.
How is this different (Score:2)
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.
common open source deep learning libraries (Score:5, Informative)
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 [berkeleyvision.org], widely used C++ deep learning framework.
Theano [deeplearning.net], widely used Python deep learning framework.
Torch [torch.ch], the software used by Google, AlphaGo and Facebook.
TensorFlow [tensorflow.org], Google's large scale machine learning framework.
CNTK [github.com], Microsoft's deep learning toolkit.
Re: (Score:2)
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.
Given the amount of vaporware announcements I've seen over the past 30+ years, that's the stupidest thing I've read this month.
Re: (Score:3)
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...?
You are totally missing the point here. By creating a FOSS based NN project, what they are actually setting in motion is a crowdsourced Genetic Algorithm (GA) that is designed to improve upon the actual NN product. Thus the NN solution you are complaining about is only the seed data for the GA process, and that this announcement is actually at the level of meta-programming the NN. You really need to come back in 10 years time to see whether the output of the GA is widely used or (most likely) is sitting
CVS? Really? (Score:2)
Re: (Score:1)
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.
Where did you find this? I was looking for the source code repo. And I agree. CVS is dooming this project from the start.
Re: (Score:2)
Re: (Score:1)
GIT isn't modern...its a tool used to torture devs, used by hipster programmers who think using the same tools as other devs isn't "cool"...now go back to making your unnecessarily complex code in a brand new framework and I'll go back to making something useful that does the same thing with 1/10 the computer's resources...
sweet! (Score:2)
Source Code Repository? (Score:1)
Hey GNU, we need open data, not code (Score:2)
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 [businessinsider.com] from its entire fleet for millions of miles traveled. Google doesn't have a fleet, so it wants to collect so much data w [mashable.com]