%0 Book %1 mcneill93 %A Mcneill, D. %A Freiberger, P. %C New York %D 1993 %I Simon and Schuster %K logic %T Fuzzy Logic

@book{mcneill93, added-at = {2006-03-24T16:34:33.000+0100}, address = {New York}, author = {Mcneill, D. and Freiberger, P.}, biburl = {https://www.bibsonomy.org/bibtex/268b05f1f688dd59d2bdb7567390e7065/neilernst}, citeulike-article-id = {121791}, comment = {Journeyman effort which sensationalizes the area of fuzzy logic without explaining it that accurately. Based on a lot of hype. Fuzzy logic vs. Bayesian reasoning: mountain out of a molehill. Bayesian is two-valued probability, whereas FL is infinitely valued. BR can model any set of solutions (e.g. it rains tomorrow, it doesn't rain, it rains half the day etc.) that FL can, but I propose that BR is more complex to describe. Think of it as describing a non-trivial quantity of something numerically. Using decimal, we can represent this in much fewer (how many?) than binary. Interesting idea was combining neural nets and fuzzy logic. The neural nets find hidden patterns and store knowledge, the FL is used to prime the system with knowledge. Maybe combine FL and regular logic? From comp.ai.fuzzy: The major problem with FL, NN, GA and all these AI approaches is that they have no "explanatory power". For example, fuzzy rules "describe" the state of knowledge about a system, but do not go beyond that- do not extend our knowledge. Take for example, quantum mechanics, electrodynamics, relativity, fluid mechanics, etc. They all have very compact laws with very few (non-linear differential) equations that help you "explain why" things happen for a huge number of physical problems. These complex equations helped us predict many new phenomena, design many new machines and expand. FL does not do that. I agree, FL is very good for non-linear control applications, in that you replace a human operator with a faster and more reliable computing machine. But can you do for example stability analysis on a FL controlled system? Can you tell a-priori what will happen when the system operates under conditions not covered by the fuzzy rules? Can you generalise the rules or extrapolate them into wider spaces and dimensions? Response: That is true. But you must understand that what we call AI are, at the very basic level, nothing more than specialized *Search* Algorithms (usually hueristic), which are maybe optimized for their specific application using some rules of thumb concerning the data on which they operate. But thats all they are: Search algorithms. They can only perform searches on the given knowledge base, and cannot really "extend" it. If they could extend it, it would no longer be "artificial" intelligence....I'd call it a pretty good form of true intelligence :)) The reason they are called "intelligent" is because they can adapt themselves to become more optimized, thus appearing to "learn". At most, they can learn to handle inputs that it was not specifically trained for, as long as there is some semblance between the expected and unexpected inputs. However these rules used to learn are coded by humans, so in the end you have to agree that its only "Artificial" Intel. If you want algorithms to actually extend your knowledge base, this is > how I'd go about it: You will have to write algorithms that capitalize ...... ...... > itself, RECOGNIZE a Pattern, IDENTIFY the major Parameters relevant to > it, and FORMULATE Equations/Data/Rules to describe it. > >...... >...... > Then you can have algos that try to find the patterns/equations in > this base, algos which try to generate new data/rules from these Some professor at UCLA (can't recall the name) once wrote that the difference between artificial Intelligence and true Intelligence is that machines don't have the "I". They don't have a self, no-motivation, no desire to interpolate or extrapolate. They won't learn unless we teach them because they have no motivation to do so.}, description = {sdasda}, interhash = {a37eeeea41d55c0757008c6016bba1ac}, intrahash = {68b05f1f688dd59d2bdb7567390e7065}, keywords = {logic}, priority = {0}, publisher = {Simon and Schuster}, timestamp = {2006-03-24T16:34:33.000+0100}, title = {Fuzzy {L}ogic}, year = 1993 }