3 ms·
Kind of odd to have this here and arguments about the accuracy using "gradient descent" vs "back propagation" both on the same HN front page. Fuzzy Logic is mos
by kurthr 6y ago
Kind of odd to have this here and arguments about the accuracy using "gradient descent" vs "back propagation" both on the same HN front page. Fuzzy Logic is mostly a marketing term, and although it may still exist other machine learning/estimation tools exist that seem to completely subsume it. If I remember correctly some fundamental aspects of fuzzy-set theory are still unproven and possibly shown undecidable.
To be honest this is more a hearkening back to the 90s retro take on The Beloved Old Fuzzy Logic of yesteryear and ignoring it's rot of irrelevance than any actual evaluation of the technology... "The smartest appliance I have ever owned". A spring and a magnet do almost as well.
- jacques_chester 6y ago> Fuzzy Logic is mostly a marketing term And it's a terrible marketing term. If it had been called "continuum logic" instead, there'd probably be much less disdain.
- djur 6y agoThe number of people I've heard mention "fuzzy logic" when lauding their Zojirushi rice cookers would suggest that it is at least a very memorable marketing term.
- jacques_chester 6y agoThey would've lauded "continuum logic" just as heartily. But everyone else sees "fuzzy logic" and immediately makes jokes or dismisses it out of hand. Names matter.
- jerf 6y ago"If I remember correctly some fundamental aspects of fuzzy-set theory are still unproven and possibly shown undecidable." If I remember correctly, and I may not, fuzzy logic fell out of mathematical interest when it was proved all fuzzy logic statements can be converted into conventional logic statements without loss, by adding some additional terms representing various ranges of values, e.g., if we have a sensor $T that senses the temperature in a range from 0.0 - 1.0, the fuzzy statement if $T > .6 then TurnOffBurner else TurnOnBurner can be converted to the conventional-logic statements $TAbove0.6 -> TurnOffBurner $TBelow0.6 -> TurnOnBurner With the obvious mapping of what the two values on the left are. Proving that the entire logic can be so converted is less trivial as the statements interact but IIRC it was done. The whole hype of fuzzy logic is, well, how it was presented in this article, that it could somehow deal with the real world better than conventional logic. The proof that there was nothing it could say that conventional logic couldn't ended that conclusively. However, it can still be a useful "programming API", for obvious reasons. But then, you don't really need a specialized "fuzzy logic" necessarily, for a lot of tasks, a big set of if statements based on non-boolean values (which, you know, are easy to come by in programming languages) does the same job. While expressing it directly in fuzzy logic might open up some mathematical analysis possibilities, expressing it in conventional programming languages opens up a lot of software engineering possibilities, so shrug. This is just another expression of the decades-long and ongoing arguments about how mathematical programming really is.
- jacques_chester 6y agoYou're probably thinking of fuzzification and defuzzification, which are engineering steps. Mathematicians have continued to make hay with the family of "monotone measures", of which probability, fuzzy logic, belief functions and so on are particular members. That I can express a problem geometrically or algebraically doesn't make one of them less useful or inspiring.
- kurthr 6y agoYes, that matches my memory... "anything you can do with fuzzy logic, I can do more generically with regular logic and a few extra states".