Fuzzy logic seemed like an active area of research in machine learning and data mining back when I was in grad school (early 2000s). Fuzzy inference systems, fuzzy c-means, fuzzy versions of the various neural network and support vector machine architectures were all being taught in grad courses and discussed in conferences.

Since I've started paying attention to ML again (~2013), Fuzzy Logic seems to have dropped off the map completely and its absence from the current ML landscape is conspicuous given all the AI hype.

Was this a case of a topic simply falling out of fashion, or was there a specific limitation of fuzzy logic and fuzzy inference that led to the topic being abandoned by researchers?

  • $\begingroup$ I believe that the closest answer is as you Alex S. King said in your question: "specific limitations"; to learn a complente response about this, please, I recommend read this link answer: ai.stackexchange.com/a/10324 $\endgroup$ Mar 3 '20 at 18:31
  • $\begingroup$ @AntonioLeonardo Unfortunately, your last edit doesn't seem to have improved your initially given answer. You were still just linking to another answer. Ideally, we're looking for more historically valid reasons that can be supported by some research work, rather than the opinion of someone else. I converted that answer to a comment above. If you have an answer with more details other than linking to another answer (that doesn't actually answer this question), feel free to add it ;) $\endgroup$
    – nbro
    Mar 4 '20 at 22:01

Fuzzy logic is not down trending. It's a common architecture selection for systems that require the representation of uncertainty in changing rules. The domains include work-flow control, aeronautics, chemical plant engineering, automated defense of cyber-attack, building systems, and business intelligence.

In natural language processing, it's up-trending under other names. Semantic nets are a logical structure with strengths assigned between nodes of the net based on meta rules, which is reminiscent of early fuzzy logic software.

For instance, the rule might be that a word that ends with -ly before a word associated with action may be an adverb modifying a verb in English, but it is not always the case. The rule's application is not smooth, meaning non-probabilistic. It is fuzzy.

There is no way to create a grammar for common speech or writing in any natural language. Language is dynamic and as practically non-deterministic as wind vectors. That's why grammar is a down-trending word among linguists and people working in NLP.

Cognition is expected to require fuzzy valuations to the black and white nature of early production systems developed at MIT, Cambridge, the U.S. Navy, Yale U, and other locations in Eurasia. People develop rules in their cortex and they rise and drop in probability when decisions are made after the rule seemed to have worked or didn't.

The logic that works to prove a math theorem may require discrete inference, but people are smart when they can weigh alternatives and apply rules in unique ways to partial data and limited experience.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.