1 / 19

What is meant by Fuzzy

4/9/2012. 2. Vision Statement. Fuzzy is taught in schools.AI scientists have made robots Which are not sound (like humans)are not complete (like humans)Do mistakes in their computations (like me)Can learn more and more (child -> Einstein).Can do what humans do.. 4/9/2012. 3. Goal and Objectiv

mariko
Download Presentation

What is meant by Fuzzy

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


    1. 4/9/2012 1 What is meant by Fuzzy Sayed Kamal Aldin Ghiathi Shirazi

    2. 4/9/2012 2 Vision Statement Fuzzy is taught in schools. AI scientists have made robots Which are not sound (like humans) are not complete (like humans) Do mistakes in their computations (like me) Can learn more and more (child -> Einstein). Can do what humans do.

    3. 4/9/2012 3 Goal and Objective Criteria: Human-like speaking Human-like speech understanding Human-like object recognition Human-like control of objects …

    4. 4/9/2012 4 Today’s Situation Speech recognizers, are unable to correctly recognize unambiguous speech. Speech recognizers, do impossible recognitions. There is a big gap between recognition rate in Databases and real speech. Has any body proved the possibility of making an ASR system, which does not mimic humans?

    5. 4/9/2012 5 Why we believe in fuzzy? Proof by contradiction. Mathematics: Godel’s Incompleteness Theorem. NP-hard Problems Physics: Relativity Theory (maximum speed of light) Quantum Theory (maximum accuracy) AI: traditional models have a fixed ability Experiment: We have done our best. No improvement is obtained. We have no hope to have a sufficiently fast computer.

    6. 4/9/2012 6 Problems with probability (1) Unable to learn. We must be aware of probability space from the beginning. Obvious difference with humans’ inference. Probabilistic logic? Humans usually learn with very few number of training examples. Experiment has shown that humans are very bad probabilistic expectation evaluators.

    7. 4/9/2012 7 Problems with probability (2) From a theoretical point of view: Bad representation of zero knowledge. Unable to change the probability space in a reasonable way. Non-symmetric deal with new and old evidence. (using conditional probability) Note: Other problems also do exist.

    8. 4/9/2012 8 Problems with probability (3) In practice: We usually define a bad probability space Good: Probabilistic Turing Machines. Bad: HMM. What about a search engine? Is it hard to make a 0.999999 percent primes detector? Zadeh’s uncertainty relation.

    9. 4/9/2012 9 Available Options (1) Shafer-Demster theory Similar to probability theory in operations. Very beautiful formulas. Few people ever know its existence. Fewer people can grasp it. Costly when we are in hurry. N.B. This theory is produced in a probability faculty only by two mans.

    10. 4/9/2012 10 Available Options (2) Possibility Theory Can join physical part of brain with its cognitive counterpart. Has lots of cognitive supporting evidences. Few people ever know its existence. Fewer people know its difference from Fuzzy.

    11. 4/9/2012 11 Available Options (3) Fuzzy Theory Has not gained any improvement over traditional methods like HMM and NN in ASR. Fuzzy Probability theory Currently is used both in HMM and NN.

    12. 4/9/2012 12 What fuzzy is not Fuzzy does not improve our accuracy by giving everything a degree of membership. Nothing (or very few things) is (are) fuzzy in syllable units. In next slides, we show fuzziness in speech.

    13. 4/9/2012 13 What … What is “n”? No, to recognize “n” in a sentence we use other sources of information to understand what a phoneme is. I want to hear the phoneme itself. this is the true way. See the two next slides.

    14. 4/9/2012 14 “i” + “N” + “A”

    15. 4/9/2012 15 “i” + “A”

    16. 4/9/2012 16 “i” + “A”

    17. 4/9/2012 17 Do current ASR systems extract the true features? See the answer in next two slides.

    18. 4/9/2012 18 “M” + “A” “N” + “A”

    19. 4/9/2012 19

    20. 4/9/2012 20 Recommendation Verify your steps in making an ASR system with that of humans. Make sure you are extracting the true features. Be accurate about values of features, but not more than humans.

More Related