Intelligence Posted August 5, 2003 Posted August 5, 2003 I am a neuroscientist and have an interest in the near future of designing a neural network. The purpose of my neural network will be to understand the structure of input sentences and to communicate in return to these sentences based on either programmed database information or based on user input information. Would anyone be interested in working on this? I am unfamiliar as to languages which are "best" for this although I do know some use C++ and that would most likely be fine. Please let me know!! My goal would be to have a website which will run this for testing. It's a simple and finite goal for NN I can do all work except the physical programming but can provide alot of information, thanks!
Intelligence Posted August 5, 2003 Author Posted August 5, 2003 I should point out a few more things. This NN does not simulate a biological neural network - it does not simulate neurons, synapse, reuptake and gradient potential gates.... ..because it doesn't need to in order to achieve it's goal. The project will largely consist of an understanding and proper dissection of a sentence into it's usage parts and labeling of each word as it's type (noun, adjective etc...) I am organzing all this myself, it's just programming which would need ot be done. ALso this isn't a paying job - it's merely a project. I plan on eventually purchasing a .com of which I have already chosen (unless it changes) I am completely unfamiliar with how the programming would go so I can't make a comment on that - I can speak further on the individual actions it would need to take so you can get an idea of what the programming would look like. Thanks all!
Sayonara Posted August 5, 2003 Posted August 5, 2003 Neural networks aren't exactly easy to program. I have severe doubts you will find anyone who would help make one out of sheer goodwill, but stranger things have happened (and of course for your needs a neural network might not actually be necessary). Have you heard of ALICE? Investigating this program and others like it might help you to determine exactly what you need.
Intelligence Posted August 5, 2003 Author Posted August 5, 2003 Yes I'm familiar with alice. Used to a free and pretty poor program. Not I think they have some other alice thing which is differnet. Alice really is merely an AI attempting to converse. There is a big different her e- mine has the purposes of understanding what was said in each sentence. As a result it would be able to converse but alice is not based ont eh same fundamental and is also not nearly as "advanced" although because mine is a different fundamental it would require LESS work to make it better. Yeah i agree it's a huge task - maybe soon I'll get someone from the computer science center at my school who is a totally programmer geek and wants to work together. It's worth a try eh!
Sayonara Posted August 5, 2003 Posted August 5, 2003 I see what you're trying to do - in this case you most likely will need a neural net because higher-order boolean decisions will probably not be sufficient to the task. It may be worth investigating the possibility of getting this set up as a sort of background project for the computer science center, so all the happy geeks can get involved in it.
Intelligence Posted August 5, 2003 Author Posted August 5, 2003 Originally posted by Sayonara³ I see what you're trying to do - in this case you most likely will need a neural net because higher-order boolean decisions will probably not be sufficient to the task. It may be worth investigating the possibility of getting this set up as a sort of background project for the computer science center, so all the happy geeks can get involved in it. Yeah, if I was to do that I would want to have a very lengthy proposal and plan based on research in books - which can be done in due time. Thanks!
atinymonkey Posted August 5, 2003 Posted August 5, 2003 Have you considered that the all conquering Microsoft probably have a legion of evil minions attempting your proposal? Just to put what your proposing into context, imagine how much development goes into Microsoft word and how relevant the helpful paperclip is.
Sayonara Posted August 5, 2003 Posted August 5, 2003 "I noticed you're writing a... something. Shall I do it for you? YES | NO."
Intelligence Posted August 5, 2003 Author Posted August 5, 2003 Monkey - not sure i understand your comment. I am not saying my idea is new at all. But the approach is unique enough to be the most widely available one. But I'm doing this for my own sake to understand ANN instead of the BNN that I deal with everyday. As far as that little paperclip - fuck that little bastard if he comes around here again I'll bend that fucking tip like a god damn twig. hate that little punk.
Intelligence Posted August 6, 2003 Author Posted August 6, 2003 Perhaps I put too much emphasis on the FINAL PRODUCT of this thing. It's main initial purpose is merely to take an input sentence and be able to give many proper values about: 1. the sentence 2. Each word in the sentence That is merely it's main initial goal This would be the input and half process phase. The second part is the putput processesing and the output itself, which is a hole nuther story. Eh you're right who the hell would wanna program this what a bore.
IMI Posted August 6, 2003 Posted August 6, 2003 Would the input be verbal or via keyboard? They already have programs that translate verbal dictation into typed text. These programs, though not perfectly yet, can also discern between such words as "hear" and "here" by looking at surrounding words for context. You might be able to integrate such a program into your interface thus saving you at least that much work.
Intelligence Posted August 6, 2003 Author Posted August 6, 2003 I'm definetely looking at it sensing textual input. I would want to completely finish this before ever integrating a speech converter - and then merely place the speech convertor on top of the textual input....... I would possibly integrate it, thanks! Right now I am researching and seeing about the project concepts and how it would be best executed...
IMI Posted August 6, 2003 Posted August 6, 2003 Web sites like Merriam-Webster online dictionary, and Ask Jeeves, already have some type of text recognition where you can input a phrase and it will search based on the words inputted and how the words are combined in relation to each other. So, the first part of your quest is already done. You would just have to modify it to give you the types of results you are looking for.
Intelligence Posted August 6, 2003 Author Posted August 6, 2003 Yes it's already done. But I don't have the technology behind it. Keep in mind I'm looking to do this for the sake of doing it. Not for the sake of showing it off or being original.
IMI Posted August 7, 2003 Posted August 7, 2003 I see. It does sound fascinating. Be careful, could lead to AI From what I know of databases I don't think it would be terribly complicated, just time consuming.
alt_f13 Posted August 19, 2003 Posted August 19, 2003 Heh, I was thinking of just this exact thing a few days ago. Would you be able to make it so the program can determine the meaning of a word by the way it is used? Much like a baby. THEN If it was able to log every "conversation" or whatever, after learning a few words, and logically determine how the word was used and then its meaning, you would have a learning program! Guess what(!) after the initial programming (might take a few versions, supposing what I'm talking about is doable or close to what you guys are on about) you could integrate the speach to text on the initial run, and *bang boom* true AI is born. Combine this with that Nissan robot and you have "The Matrix: Preloaded"
Skye Posted August 19, 2003 Posted August 19, 2003 Anyone here actually tried to create a neural net?
Dave Posted August 19, 2003 Posted August 19, 2003 If you have a look on http://freshmeat.net/ you'll find a neural network library that's written in C++. I think it's called libneural or something. You basically feed it test data, which creates a bias. Anything over (or under, I can't remember) that value fed to the 'neuron' causes it to fire. Beyond this, I don't really know that much about it.
Guest airhead Posted August 20, 2003 Posted August 20, 2003 To the first guy who posted this thread: Can you give us a little of your credential? Why do you think this would be worthwhile doing? Several people have used neural nets to try to understand language process -- some have been shot down by others. Anyway, I didn't realize this is what neuroscientists do -- always thought it's more of a linguist's job than anything else. What can I say.. I am just an airhead. To the other guys: Neural networks are actually really simple to code -- in the naive implementation. Neural networks in general is such a large field. How do you plan to represent your data? There are questions that this guy hasn't provided and probably hasn't worked out yet suggesting that he hasn't done his research in this area. In fact, there has been lots of efforts to use neural nets to process language. Here's a brief overview I found just by googling up some words: http://www.hemming.se/gslt/LingRes/NeuralNetworks.htm Lots of detailed works can be found in more specialized journals.
m@ Posted September 28, 2003 Posted September 28, 2003 atinymonkey said in post #7 :Have you considered that the all conquering Microsoft probably have a legion of evil minions attempting your proposal? Just to put what your proposing into context, imagine how much development goes into Microsoft word and how relevant the helpful paperclip is. Open office is for free on linux and it has as many if not more features than microsoft's word
YT2095 Posted September 28, 2003 Posted September 28, 2003 So how do NN`s differ from Boolean based nets? I have a little experience with neural cybernetics and cellular automata type AI, and other thatn needing vast amount of memory set up as dimension arrays, the core or rather "Seed" program is just that, external input or "Experience" modifies the original seed prog, and so after many runs the seed is nothing like the original, but ALWAYS still just a program? and can also be tracked as to its modifications with a simple tracer log (there`s nothing mystical involved, including emergent behaviour)... so how do NN`s differ? coz the stuff I was doing was out in the late 70`s, I did mine in the early 90`s, only difference was the price of mem and chips were cheaper, so bigger and faster ones could be made for a fraction of the cost. don`t you need biological stuff for the sort of thing he`s thinking about? like when they grow neurons on a chip and all that stuff? I say good luck to him
Pinch Paxton Posted November 11, 2003 Posted November 11, 2003 You would want to look through a sentence, find the spaces, chop out the individual words, store their positions in the sentence, and compare them to a lingustic tree. Ask questions about the answers to build up a bigger tree. The NN part of the program would be the part that builds the tree. That works as Yes/No/Maybe, and stores the words, and replies based on those three questions. If you just want to understand it, then you should program it yourself. I don't see how you can understand it if someone else programs it. Pincho.
Guest sim Posted January 12, 2004 Posted January 12, 2004 maybe he just wants to view the output based on his theory of programming it. original poster, id be interested in helping. i seen someone propuse an idea similar to this befor but everyone declined and i didnt like the way he wanted it done. i dont understand your whole theory on this, but from what im gathering you want something to start with figuring out what each word is. it seems that you would actually need a readable database as a dictionary which i couldnt seem to find. telling whether its a noun verb, ect. then the definition of the word. i even email a few people asking them ot make one for me, but prices were way to high 10,000$.
mossoi Posted January 13, 2004 Posted January 13, 2004 Hmm - this sounds interesting and if I'm get what you're looking for it's not a neural net that you're trying to create, you are just modelling something that can follow a programmed set of rules in order to perform an action based on the input. This is quite easy to do and can be implemented as a web app. allowing access for several developers and possibly even human trainers to get through the leg work of data input. I might be interested in this as a project depending on your time scales etc. Let me know if you're still interested in doing this. You can get hold of me through my site, http://www.lobsterweb.com, if I'm slow to respond here.
Recommended Posts