Quote: Original post by Timkin
I used to be quite anti-FL, mostly because of my Bayesian background and training in probabilistic AI. However, I've since come to understand that FL definitely has its place (albeit a limited one, much as ANNs). It was the realisation that FL is equivalent to a B-spline representation of the domain (using triangular basis splines) that brought the most insight (for me) into its appropriate use.

At least as far as AI is concerned, one should not treat FL as a means of representing uncertainty in knowledge about a domain (for which probabilities are very appropriate) but rather as a means of representing vagueness about the actual attributes of the domain upon which knowledge is recorded.

Cheers,

Timkin

I think an easier way to sum it up is like this:

example: A weatherman says there is a 50% chance of rain tommorow. That is a probability. The next day it rains, and the weatherman says "the rain was mild".
That is fuzzy.

Quote: Original post by Anonymous Poster
I think an easier way to sum it up is like this:

example: A weatherman says there is a 50% chance of rain tommorow. That is a probability. The next day it rains, and the weatherman says "the rain was mild".
That is fuzzy.

Hehe... I see that as an example rather than a summary! ;) A good example though that I agree, captures the essence of the difference in appropriate usage.

Cheers,

Timkin

"Fuzzy state machines are very fun. Dissimilar to Probabilistic but kind of appears like it."

That's the point in my opinion.
Fuzzy is actually a deterministic technicque but it should produce some kind of probabilistic behaviour, but...this this the key point..preventing our character from doing "non logic" actions.
The question is, is it true ?
I coded a simple probabilistic FSM and a fuzzy logic algo
Honestly I did not notice any difference, so I went for the simple FSM.
I would like to know the opinion of others programmers.

Quote: Original post by AlbertoT
I coded a simple probabilistic FSM and a fuzzy logic algo
Honestly I did not notice any difference, so I went for the simple FSM.
I would like to know the opinion of others programmers.

It's fairly trivial to code a probabilistic state machine (also known as a stochastic map) to behave as an arbitrary fuzzy mapping. It's not so easy to code a fuzzy mapping to act as an arbitrary probabilistic FSM. Consider the case of modelling p(y|x) given discrete x and continuous y. FL can handle the reverse case easily (x continuous and y discrete) but not the other way around.

That's the essential difference. There's also some mathematical issues separating the two, but that's not overly relevant to this discussion.

Cheers,

Timkin

Hello

coded a "chasing" algo.

The input variables were distance and deltaDistance
The fuzzy variables were close, near, far...etc
Honestly I was expecting a much more realistic behaviour then using a simple
"if distance then ..."
I mean "hard to predict but logic "
Honestly I did not note any difference
Maybe I did something wrong
I womder wether someone made similar comparison tests