fuzzy logic [1227.2005]

I couldn't sleep last night, so I read one of the books I brought with me. It's a book called Fuzzy Logic for Business and Industry and it may very well be the worst bok I have ever read. Not only was it full of obvious typos, it just didn't make a lot of sense.

For example, there's a whole chapter on how fuzzy logic can accomodate conflicting expert opinions. It sounds like a great feature on paper, but if you think about it, a much more sensible approach is to explore the reasons for the conflict and make a decision. But what fuzzy logic does is more like a compromise.

The main problem with the book was that it was never really clear to me what problem the examples were trying to solve. This is supposed to be a book on applications, but instead it reads like a book on technology with a lot of weak examples. There was no real problem statement, and no evaluation of whether the fuzzy solution solved the problem.

Well, whatever. There are a handful things that got me interested in fuzzy logic again.

One is this script I havet that watches over my server. It's a control system, and fuzzy logic is good for that kind of thing. Instead of "if load > 2 and process.age > cutoff then process.niceness = 10", you can say "if load is high and process is old, increase niceness" and the system can figure out exactly how nice to make the process. (Nice processes are given lower priority by the operating system)

The second application I'm interested in is for my project list. It's hard to put a number to things like effort and risk and importance. The book talks about using fuzzy labels to query a database full of crisp data, but I'm more interested in storing the fuzzy values directly in the database. I'm not 100% sure what that means yet: maybe some kind of visual representation of how important rather than just a number. I'm not sure.

The third application I'm interested in is just for fun, and came after listening to On Intelligence on the way down here. On Intelligence is a book on neuroscience and machine intelligence by the guy that invented the palm pilot. Now that was a great book.

On Intelligence has nothing to do with Fuzzy Logic. More like very powerful and large neural networks. But one feature of the system he describes is the ability to recall a pattern given a patial version of that pattern. Neural networks can do that, and I think some fuzzy logic systems may behave that way (Fuzzy Associative Memories). I didn't get a clear answer from the book, but FAMs seem interesting on their own.

I've got a couple books on Fuzzy logic at home, including a more recent book by this same guy. I hope it's better than the old one...