| |
| |
| |
|
Page: 1 2 3 4
Comments:
<0> gya much
<0> gay*
<1> hah.
<1> DanF_DrC: ANy suggestion reading? I finished GEB and am looking for some more material =)
<0> GEB?
<1> godel escher back, mainly about formal systems although it had a bit on AI
<1> wow, my spelling/grammar is wretched this evening. bach
<0> neural nets. a few pages in Mitchell's 'Machine Learning'. can be found online on a mexican site in the topic link
<0> also pomerleau paper on project ALVINN which is a simple but instructive example of real world application of NN
<0> NN for the win
<1> yeah I've read all those. I had a "Fundamentals on Neural Nets" book i've gone through a while ago
<1> =)
<0> ok then you could train neural nets with input as both input and output and have a smaller middle layer to get automatic feature discovery there. realize the importance of that and build scifi level androids
<0> unsupervised learning
<1> haha
<1> everything sounds fine except that whole "automatic feature discovery"
<0> how so?
<1> Ah sweet, that mexican ebook site has the russel and norvig book =)
<0> that's a piece of crap but who's counting
<1> DanF_DrC: So you would recommend the mitchell book over that?
<0> look at the front page illustration. that's what we call misunderstood subject and receives the lowest grade possible which in Denmark used to be 00 where 6 was needed to p***
<2> that's right the smaller middle layer with automatic feature discovery
<3> poolio: are you looking for technical intro, or philosophical?
<1> yaroslavvb: I'm up to really either. I just read GEB, and On Intelligence and am either going to continue reading or start up a project involving something or other.
<1> What I'm really interested in having the ability to do is creating a robot that through unsupervised learning is able to develop its own gait and such
<0> that's quite possible and the right way. just need to think about it
<0> yaroslavvb, still in the bayesian camp?
<0> he never answers
<1> DanF_DrC: Any links on "automatic feature discovery" ? Don't know anything about it
<0> it's another word for unsupervised learning
<0> to exemplify it's the ability to recognize a pattern like lines in an image with being told they are lines
<0> without*
<1> So the approach that you believe is that you have a recurrent NN which has an intermediate layer which attempts to extract features of the dataset?
<0> not really recurrent no
<0> recurrent as I understand it is cyclic right?
<0> as in the signal runs round and round
<4> poolio: Feature discovery is a very difficult task. What little research I've found isn't very advanced, or readable.
<1> yeah
<0> it's not recurrent. just the input as both input and output and train with backprop
<0> that might seem odd but it works
<0> with a smaller middle layer you force it to interpret the data
<4> If the input is the output, then it's recurrent...
<0> and it's that interpretation that is golden
<0> sigh
<0> it's a way to attain understanding automatically. if that can't be grasped then you'll never be able to do it
<1> Yeah... The problem is I have no clue whatsoever how to approach a feature-extraction problem. I was thinking about looking into Machine Learning and Pattern Recognition, the new one by bishop
<2> should this kind of thing be your first exposure to AI?
<0> you don't understand. you could read all the books in the world and not one will tell you something more valuable than I just did
<0> on AI that is
<2> is a significant amount of prior knowledge required to do this?
<0> the bible is another matter
<2> or is it mostly mathematical?
<1> DanF_DrC: Yes, but I feel that understanding what currently works/doesn't and the current approach some people have had will benefit my own research in the future.
<0> no neural nets are rather simple in concept and require very little prior knowledge
<2> does a neural network have to be spiking to be recurrent?
<0> poolio, I can't understand it for you but what I'm trying to tell you is that nobody has done anything that works at all. you might take a look at TD-gammon but that's quite irrelevant compared to what I just suggested
<0> yaroslavvb, no answer? funny how that works
<1> DanF_DrC: Yeah. I understand the gravity of the problem, I was just wondering about some statistical approaches to finding similarities/patterns in data sets.
<0> no you don't understand. I'm telling you that it's easy
<4> Psaronius: No, a network is recurrent if it's network structure is cyclic. A network is spiking if nodes values are updated asynchronously.
<1> If it's so easy then why has no one done it?
<1> DanF_DrC: Do you believe that more closely modeling an intelligent being will increase the likelihood of creating an intelligent agent?
<4> poolio: I'd take Dan's advice with a grain of salt. He enjoys talking about AI far more than researching it.
<1> Cerin: I take everything on the internet with a grain of salt, a couple pills, and some water =)
<0> well most people are morons. just look at yaroslavvb who can't even answer a single question because he'd rather close his eyes forever than admit he was wrong. And I don't do it for ethical reasons
<4> heh, wise move :)
<0> and even though I offer the solution in detail nobody is able to appreciate its importance. so be it :)
<4> DanF_DrC: yaroslavvb probably has you on /ignore, as do most people
<1> Cerin: any suggested reading?
<4> poolio: Probably nothing you've already read. I find most books are behind the times, so I try reading what I can find an citeseer.
<4> *on
<1> so mainly reading current/very used research papers?
<0> Cerin, no he doesn't. you try asking him if he is still in the bayesian camp
<1> Cerin: I'm a noobie to AI =P
<4> poolio: yeah. They can be difficult to read some times, but at least they're fairly recent.
<4> I finally got around to looking at Machine Learning, and was disappointed to see the only reinforcement learning algorithms it covered was Q-Learning.
<0> Cerin, no?
<5> anybody some thoughts about 'Artificial General Intelligence' by Ben Goertzel and C***io Pennachin?
<0> haven't read it but it's no doubt crap
<5> no doubt?
<4> poolio: I've been reading some papers on algorithms like DSHP and HEXQ, that leave Q-Learning in the dust.
<5> DanF, why would it be crap?
<0> BlackHwk, ever seen 'artificial general intelligence' used anywhere?
<0> seen any droids in the streets?
<4> poolio: And don't feel bad about being a newbie. It's a young field. The brightest minds researching this stuff are basically newbies as well ;)
<1> Cerin: Yeah, I'm debating between majoring in CS/PHI or Biomedical engineering
<5> DanF: ever seen quantum mechanics used anywhere before 1900? don't see why a new term automatically would be crap at all
<1> Cerin: So in terms of unsupervised learning, any good intro readings you suggest? I've done some basic NN implementations but that's about it
<4> poolio: I'd go Bio. That's where all the money and hotest research is coming from.
<1> yeah, that's what i keep hearing =P
<4> poolio: Sorry, can't recommend anything.
<0> BlackHwk, maybe they would use it for something before writing an entier book about it
<0> entire*
<1> Ah no problem. Any opinion on the new Bishop book?
<5> sorry DanF...if that's all your reasons for your opinion...;). Anybody else know the book?
<4> No, I'll have to give it a look.
<0> you have such blind faith in authority that you can't believe that nobody knows **** that you keep looking for new books forever rather than stand on own two feet
<2> I am also taking the biology approach. It seems to me that whenever you have a beginning field, it's best to take examples from something that's preexisting rather than trying to reinvent the wheel
<1> YES
<1> DanF_DrC: It's published, it must be brilliant!
<0> hehe
<0> yeah right
<0> how naive is that
<5> hehe
<4> heh, the college textbook field is such a racket
<1> Sarcasm and the internet don't agree.
<0> you could find noone in the world less clueless than I with regards to AI. none
<5> hey...hypothetical situation: computer scientist with a dream to some day built an intelligent computer about to do his thesis. What should he do?
<5> ;)
<1> Commit suicide?
<4> BlackHwk: Solve the challenging game of tic-tac-toe ;)
<5> hehe, something one could finish in a year maybe? ;)
<4> I was thinking more of a single day ;)
<0> BlackHwk, look up the requirements
<4> Maybe two days if you want to implement a fancy alpha-beta engine.
<0> the parameters
<0> in Denmark we're not allowed to think on our own for a master's thesis. only study other's work and master that
<4> You're not allowed to think until you do your PhD. thesis.
<5> ah ok. pretty sure i can make something up myself...but still has to be approved
<4> BlackHwk: What are you interested in?
<5> problem is that there is so much stuff out there! :s
<0> I made the 'mistake' of thinking something brilliant up and it was not appreciated. at all
<5> and you're not a millionair yet?
<4> Ah, true artists are never appreciated in their own lifetime! :)
<1> DanF_DrC: I enjoyed that qspan paper you did a little while ago. that's something i'd like to be able to make happen
<5> natural language processing is mainly what i want to do. see how far an intelligence can get with just distributional information
<0> that was preliminary to my masters degree and they failed me for it :)
<0> granted the supervising professor was an idiot from the bayesian camp (redundant term) but still
<5> and if possible i would like to use an evolutionary approach in which the actual cognitive machinery is evolved. Big problem in that case is to apply/find the right selective pressures
<1> BlackHwk: there was some cool paper i read on using an SOM for phonemes
<0> the whole department is infested with bayesian and has been for 35 years so a little punk coming along insisting on using neural nets.. wasn't appreciated
<0> so I had to go
<0> not unlike yaroslavvb the blind's behavior
<5> poolio: Creutz & Lagus?
<1> DanF_DrC: No other places you could go with your work?
<4> BlackHwk: Yeah, that's an area I've been interested myself. Unfortunately, few people have any idea on where to begin. Intelligent responses need a lot more than just a statistical model of a language.
<1> BlackHwk: Have no remembrance
<0> poolio, perhaps but with the breakthrough and the ethical considerations there was no need anyway
<1> DanF_DrC: So what do you do now?
<4> yeah, DanF_DrC is the archetypal example of a trauma victim. His brilliance was spurned back in college, and now he extracts his wraith against all those he ***ociates with those who spurned him.
<0> computer graphics and the occasional software dev
<5> http://www.cis.hut.fi/AKRR05/papers/ << paper 14
<0> hehe
<5> hehe
<1> not that one, but similar idea.
<0> I think you will find that I just kept growing
<5> Cerin: happen to have any idea on what focus research is currently taking with respect to these other factors that are needed for intelligent behaviour besides the statistical model of language?
<0> but of course you have your great success in AI to boost your position on this..
<0> it's very curious how noone can be elevated in this matter. no matter how spoon fed I make it you always revert
Return to
#ai
or
Go to some related
logs:
dolmans recipe
kaudiocreator sg_write
HYPOTONEUSE,TRIANGLE FORMULA
#web
fakeraid centos
debian wifi accesspoint
setting Mix_OpenAudio to 4 channels
Telnet to the bot and enter 'NEW' as your nickname.
#openssh
#centos
|
|