Need your input on the topic "Neural Networks in different programming languages"

lvomrm · 2012-05-26T17:15:00.233136+00:00

artificial consciousness? But your not telling me, that there is a real AI, right? So something that has real original thoughts and not reproductions of the makers intelligence, right?

lvomrm · 1 day

I think compilator sound a lot better, a little like some super-villain.

EA are something I am interested in, actually I have a project which tries to combine NNs with EA in a way.

lvomrm · 1 day

If you are interested take a look at the other thread over at /r/machinelearning. There is a little more going on (link is in my original post).

lvomrm · 2 days

Good thinking, althogh the task of writing a compiler sound quite monolithic to me (never written a compiler before). Defeniteley sounds not like a task you could achieve in 3-4 months (with normal uni-workload).

But I will look into it.

lvomrm · 2 days

Also sounds good, espacially as it seems a little more modern and seems to have more real-life use. But that might be exactly what that professor would not want to see me doing. I also have to work around the fact that my university is lacking people who are rally into machine learning.

To clarify this for me, MachineLearning is not just NN-Work, right? I'm not yet involved that much, but when you say MachineLearning you also mean other mechanics, right?

lvomrm · 2 days

This sounds so straight forward that it sounds like sarcasm. But looking at up- and downvotes I guess you are serious.

Thanks for the opinion! It don't think it will be a breakthrough, but I would really like to do something that has anything to do with neural networks (and/or AI), and as all the profs at my university who where in some way doing somehting in that direction are retired, this might be my only chance...

lvomrm · 2 days

Ok, so that would suggest it is more usefull to examine those few libraries and compare them.

But I thought about feature on the higher level of programming languages. HELOSMTP mentioned a few in his comment earlyer.

lvomrm · 2 days

Thank you, it does help. And hinder ;)

On one site I know am convinced that pretty interesting research could be done in that direction. It seems I could really find some "fundamental" (not weighing this word too much...) facts there.

On the other hand... it sound like it will be some time till I have the knowledge to do so, at least at a large scale. For now I would have to find some sub-topic of this big bunch, otherwise it will be way to big for me. I'm not sure how the american system works, but in germany this would be my bachelor-thesis, so after 3 years of studying. And as far as my Professors are concerned I didn't even hear the phrase "neural network" on campus.

lvomrm · 2 days

That makes me think that this would be an interestint topic to explore... but still not sure if it's the right one for my thesis.

lvomrm · 2 days

Sounds interesting, the inspirations leading to Io are nearly a selection of languages I always wanted to look into :)

lvomrm · 2 days

There are a variety of different ways to implement neural networks even in a single language. An external library can be used, or a native implementation that may or may not make use of threading, vector optimizations, GPU computing (OpenCL, CUDA), etc.

Those where exactly the differences I was thinking about. And it really sounds like a very big topic for one thesis.

Given all of the factors to consider, the question you are asking is not very straightforward.

That's where I needed you guys to help me :) It's all very rough-shaped at the moment.

Perhaps you could do just a comparison of the different neural network libraries out there.

Do you mean pick one language and compare the libraries provided for them or pick a selection of libraries without regard to the laguage? Both sound like a possible way to narrow the topic to a manageable size.

lvomrm · 2 days

Thanks for that link, that looks interesting in any case :)

Sorry, but what is an EA? (I thought while we are on the topic I ask rather than google it...)

maybe you could look into using neural networks ... for optimizing compilators? learning what optimizations are good tradeoffs etc.

I'm not sure I get what you mean. Did you mean "compiler" (or is compilator just a word for compiler I don't know?)?

And do you mean how compilers can optimized NN-implementations or if you could use NN to optimize compilers for another language?

lvomrm · 2 days

Doesn't this basically boil down to how fast you can do matrix multiplications and update the weights? Or how easily you can interface with something (BLAS, MKL, Eigen etc.) that does?

That's where my lack of experience (at this moment) becomes a problem. I imagined that there are different ways to build networks in different languages, for example using object-oriented languages to have nodes as objects (which can act more independently) opposed to fast functional processes when having a matrix. possibly also building a user-friendly network-builder (with, like, buttons and blinking lights ;) ).

lvomrm · 2 days

Do you not have some sort of supervisor for your thesis to suggest this idea?

Problem is, our university does not have someone who is working on neural networks. That's why I would be doing this with a professor that specializes on Programming Languages (i.e. building compilers and developing new languages). He told me if I can make the topic (neural networks) to get a language-angle he would let me do that.

Would it not require you to be able to understand a lot of types of programming languages in depth and then have to implement a range of different neural networks in all these languages.

That's what worries me, that I will have to get comfortable with a lot of languages and that sound like a huge workload.

put_in_da_trash · 3 days

Agreed.... Give me back my 5 minutes! I could have read something interesting....

Reptar_Attack · 5 days

Did you seriously think you would get real information about one of the mosst powerfull organisation in the world in a lousy reddit-channel? We don't work that way.

^{^{^{^{^{^{^{^{^{^{^{^{^{^fnord}}}}}}}}}}}}}

lvomrm · 8 days

Alright, thanks for playing :)

Careful_Houndoom · 9 days

I also use eneloops, but don't use them very much. And they are great at that too, they don't loose (a lot of) charge while lying around. Other rechargebles to !

Inji · 9 days

Hi.

First of all: Great Stuff! I'm looking forward to browse in alexandria!

Also: I saw that you want to use DropBox as one of the Hosting-sites and told people to ask you if they don't have an account. Some time ago I made a referral-Link list to pass to people who want a new account and want to have the free 500mb. As both partys have an advantage I thought that was a good Idea. If you want to pass the list, here it is

Furiousturtle · 10 days

Of course religion contains claims. That is obvious.

What exactly is the standing question? Do you still believe that if you disprove one claim, it does not matter at all to the whole construct? Please believe me when I say: Religion is not subject to the concept of wrong and right!

Furiousturtle · 10 days

What you describe is not fuzzy, but Religion does not work like that. Christianity is not just a set of claims! That's where the emergence comes in. A religion is more than the sum of its parts. if a religion was nothing more then a set of claims, sure, you could disprove it.

The point is that there are thing which are not that straight forward. Maybe our minds are just too limited at this point, maybe it is absolutely not possible to grasp the whole thing. I don't know.

lvomrm

TROPHY CASE

you'll need to login or register to do that

create a new account

login