Sign in to follow this  
darkm00n

Bayesian Artificial Intelligence

Recommended Posts

Hey all, I posted on here a few months back asking about a cluedo game based project I was doing. I got the recommendation to focus on Bayesian Networks instead of using what I was originally, Neural Networks. I bought 'Bayesian Artifical Intelligence' by Korn and Nicholson (it's a great book!) and have tried to read up as much about Bayesian Intelligence as possible. I've read up to chapter 4, Decision Networks and I think I understand it and how to implement the algorithms. But I still don't understand how to apply it, Bayesian Decision Networks, to my problem - I'm not sure about the causal implications/independence relations and the structure. In the book they use a Poker example, and I assume that it'd be similar as it uses a bluffing (utilises information that is implied or "guess-timated") technique but it doesn't really use causal logic (as far as I can see). Does Bayesian Logic need to be causal or can you set up any network based on any wacky logic you choose to come up wit? Thanks again for your help (if I get any), Darkm00n :)

Share this post


Link to post
Share on other sites
All I know about this subject is that caml-list uses Bayesian filter to block spam, and has somehow decided that the words 'Brandon' and 'Seattle' are strongly correlated with spam. In reality, my posts are probably strongly correlated with "things that have not typically appeared on the mailing list before." I'm not, after all, a language wonk PhD and don't post on that sort of thing. So now there's this wonderful normative effect, that because I'm often blocked by this mindless idiot Bayesian filter, that it's somehow making a correct 'moral' decision about the content of my posts! Or at least my detractors love to yabber about it. Leading to an extremely sarcastic addition to my anti-Bayesian .sig for that forum, "Is my technical content showing?"

So what I'm saying is, maybe you should consider what a dumbass Bayesian anything can be, before investing tons of time in it or whatever. I'd love to hear other people's rants about Bayesian anything they've run into.

I fear for the society that starts using these things as regular cultural tools. Can you imagine if someone started using these filters for anti-terrorism or credit card fraud? What a stiff business for the ACLU that will be! "The right not to be prosecuted by a dumb computer."

Share this post


Link to post
Share on other sites
Quote:
Original post by darkm00n
Ahhh.. vanevery0 thanks for your less than creative input/rant. Anyone else have any ideas? :)


You're welcome! I'm just making a Wild Assed Guess that the anecdote might have resonance for you someday, possibly when you're debugging your Bayesian whatever...

Share this post


Link to post
Share on other sites
Quote:
Original post by vanevery0
So what I'm saying is, maybe you should consider what a dumbass Bayesian anything can be, before investing tons of time in it or whatever. I'd love to hear other people's rants about Bayesian anything they've run into.
This one time, I was using a toaster oven to paint my house with, and it got all these nasty streaks on my walls. So now I never cook anything with toaster ovens. They're obviously a dumbass technology. What if someone decided to use toaster ovens for credit card checks? They'd destroy the world!

Share this post


Link to post
Share on other sites
Okay folks, lets get this back to the original question, which was NOT 'what is wrong with spam filters' or 'why X hates Bayesian Inference'. Please stick to the question as posted:
Quote:

"Does Bayesian Logic need to be causal or can you set up any network based on any wacky logic you choose to come up with?


(btw, the answer is no, your network does not have to be causal).

Timkin

Share this post


Link to post
Share on other sites
I have another question for you:

I've decided to use a DDN (Dynamic Decision Network), based on time, and I have a fair design for the network but I've realised that in Cluedo, if you ask a question with a weapon/person/place combination and ask the same person over and over again, wouldn't you get the same answer back over and over again?

Wouldn't you need to remember past card combinations.. say you asked a person a question and they had none of the cards in the question, wouldn't you need to remember that?

If so, this means I need 2 sets of parameters for each DDN,
(because i'm going to have 1 DDN for each set of cards (weapons, places, people) to reduce complexity), the weapon/places/people parameters and previous person parameters? Hmm.. confusing..

Anyone know how to represent this? I have a poker game example (Korb and Nicholson), but it consists of 4 rounds with 4 different Bayesian Networks, but the number of rounds in cluedo is usually 10 plus, making for a larger network using this approach,

Any help would be appreciated,
Thanks

[Edited by - darkm00n on October 9, 2004 1:43:03 AM]

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

Sign in to follow this