Stephany CoffmanWolph PhD West Virginia University Institute of Technology WVU Tech Assistant Professor Dept of Computer Science amp Information Systems Quick Intro on Me Assistant Professor WVU Tech Department of Computer Science and Information Systems ID: 577622
Download Presentation The PPT/PDF document "Artificial Intelligence (AI): Trying to ..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Artificial Intelligence (AI): Trying to Get Computers to Think Like Us
Stephany Coffman-Wolph, PhDWest Virginia University Institute of Technology (WVU Tech)Assistant Professor, Dept. of Computer Science & Information SystemsSlide2
Quick Intro on Me
Assistant Professor, WVU Tech, Department of Computer Science and Information SystemsFounding member and current co-faculty advisor of AWESOME (Association of Women Engineers, Scientists, Or Mathematician Empowerment)
PhD from Western Michigan University, Kalamazoo MI
MS from Bowling Green State University, Bowling Green OH
BSE from University of Michigan, Ann Arbor MI
Dissertation Title:
“
Fuzzy Search Strategy Generation for Adversarial Systems Using Fuzzy Process Particle Swarm Optimization, Fuzzy Patterns, and a Hunch Factor
”
Master
’
s Project:
“
Predicting Future Class Enrollment Using Neural Networks and other Methods
”Slide3
Abstract
AI is a field older than most realize – the term was coined in the mid 1950s. The field is comprised of many subfields but the main focus is on building intelligent entities. In order to achieve this goal many subcomponents need to be built, including methods for assisting computers to think like humans. Fuzzy logic is one mathematical method of doing so.
This talk will briefly introduce both AI and fuzzy logic, as well as discuss the application of fuzzy logic into algorithms to encompass more human-like decision processes. Adding a
“
hunch-like
”
element can further enhance an algorithm
’
s ability to mimic human decision making.Slide4
What I Am Going to Talk About?
Brief Introduction to AI and the Turing TestBrief Introduction to Fuzzy LogicFramework for Creating Fuzzy Algorithms
The Hunch Factor
Applications of Fuzzy Algorithms and Future WorkSlide5
The First Electronic Computers
Created in the late 1930s and early 1940s during World War 2 (Pictured ENIAC)Slide6
Artificial Intelligence (AI)Slide7
Artificial Intelligence
Movies and TV ShowsSlide8
Facts on AI
Artificial Intelligence is a multifaceted field focused on mimicking human behavior/mannerisms using computer algorithmsAI attempts to both understand how we think and how to build intelligent entities
AI became a field soon after World War II
The term
“
Artificial Intelligence
”
was coined in the 1950sSlide9
4 Categories of Artificial Intelligence
Thinking Humanly
Thinking Rationally
Acting Humanly
Acting RationallySlide10
Alan Turing
Subject of the Film: The Imitation Game & Played by Benedict CumberbatchSlide11
Acting Humanly: The Turing Test
Proposed by Alan Turing in the 1950sDesigned to provide an operational definition of machine intelligenceThe basics of the test:
A human interrogator poses written questions. If the interrogator cannot distinguish between the written responses of a human or a computer - then we consider the computer to be intelligentSlide12
The Total Turing Test
Note: the original avoided physical simulation of a humanThis version includes a video signal so that the interrogator can test the subject’
s perceptual abilities
Also this version allows the interrogator to
“
give
”
the test subject physical objectsSlide13
The 6 Disciplines of AI
Between the two Turing Test versions, it was determined that to be considered intelligent, the computer would need 6 capabilities - each became a discipline/area of AINatural language processing: ability to communicate
Knowledge representation: store what it knows/hears
Automated reasoning: use stored information to answer questions and draw new conclusions
Machine learning: adapt to new circumstances and detect/extrapolate patterns
Computer vision: perceive objects
Robotics: manipulate objects and moveSlide14
The 6 Disciplines of AI
My personal research focuses on 3 of the 6Natural language processing: ability to communicate
Knowledge representation: store what it knows/hears
Automated reasoning: use stored information to answer questions and draw new conclusions
Machine learning: adapt to new circumstances and detect/extrapolate patterns
Computer vision: perceive objects
Robotics: manipulate objects and moveSlide15
What is Fuzzy Logic?Slide16
Fuzzy Logic
Introduced in the 1960s by L. Zadeh as an expansion of Boolean logicSlide17
Fuzzy Logic Defined
Extremely popular in control systems (e.g., toasters, high-speed train controls, camera filters)A set of rules and techniques for dealing with logic beyond a two-value (yes/no, on/off, true/false) system
An abstraction of two-value logic designed to mimic a more human like approach to decision making
Allows for not only multiple values but also an overlap of values between fuzzy setsSlide18
Fuzzy Logic
=Degrees or Ranges of ValuesSlide19
Membership Functions
Membership functions are used to represent the degree of membership an element has to a Fuzzy set and the values range from 0 to 1.
0 represents not in the set
1 represents entirely in the set
Numbers between 0 and 1 represent some level of being part of the setSlide20
FuzzificationSlide21
Fuzzification
A method of adding abstraction to data, operators, or a conceptFuzzification of data is the process of taking
“
raw
”
/non-fuzzy data and converting it into fuzzy data
Fuzzification of operators is the process of converting a mathematical, logical, or comparative operator to its fuzzy counterpart, which operates on fuzzy sets instead of pure numbers
Fuzzification of concepts, the most difficult of the three, is the conversion of an idea into a corresponding fuzzy version
These three techniques, together, can be used within my framework to create a fuzzy algorithmSlide22
Fuzzy AlgorithmSlide23
Fuzzy Algorithm
A fuzzy algorithm goes beyond simply fuzzy data and includes fuzzified operators and/or concepts within the algorithmIt is trickier to determine if the algorithm is fuzzy when only the operators have been fuzzified because of the fine line between an operator and a fuzzy operator – even when operating on fuzzy data
A fuzzy operator is differentiated from the basic operator by being re-written to accommodate the meaning of the fuzzy data
A fuzzy operator operates on fuzzy sets (not straight numbers)
When the algorithm is modified to include a fuzzy concept, the algorithm is undeniable a fuzzy algorithmSlide24
Fuzzification of Concept
Fuzzification of a concept is the most challenging of the 3 presented mainly because a concept can be difficult to defineConcept = an essential element from the algorithm
Like the fuzzification of data or an operator, the purpose is to create an abstract version of the non-fuzzy
“
concept
”
Fuzzification of concepts allows for more human-like decisions and algorithm processing
Humans fuzzify concepts without even thinking about it. The human brain categorizes & sees connections between items easily. (Computers fundamentally operate on a No/Yes, False/True, 0/1 level)Slide25
Example: Fuzzification of Sorting
A common algorithm is the sorting of a list into alphabetical or numerical orderExample Grocery List: bananas, crackers, grapes, potatoes, cheese, apples, pretzels, and powdered sugar
Sorted Alphabetical: apples, bananas, cheese, crackers, grapes, potatoes, powdered sugar, and pretzels
We can fuzzify the concept of a sorted list by defining a fuzzy sorted list:
Contains all the elements in the original list (no lost of information)
Sorting is not completely alphabetical – just by the first letter (i.e., grouping the items together by the 26 letters in the alphabet)
Fuzzy Sorted List: apples, bananas, crackers, cheese, grapes, potatoes, pretzels, and powdered sugarSlide26
The Hunch FactorSlide27
The Hunch Factor
Mathematical attempt at mimicking the human internal intuitive decision process
Supplies a human
“
hunch-like
”
element into the decision-making processes of an algorithm
Acts as a fuzzy learning component and is continually altered during algorithm execution
Also provides memory for the algorithm
The Hunch Factor is stored as a fuzzy value and, thus, represented by a membership functionSlide28
The Hunch Factor: How It Works
Guessing well → the hunch factor
“
encourages
”
continued behaviors
Guessing poorly → the hunch factor
“
influences
”
the system to try a different area of the solution space
The hunch Factor changes based on observed information and can be manipulated in various ways
Height Increase, Width Decrease → More Confidence
Height decrease, Width Increase → Less ConfidenceSlide29
Applications and Future WorkSlide30
Successful Applications:
Fuzzy Process Particle Swarm Optimization (FP2SO) (with and without the hunch)Fuzzy Patterns Based Approach for Environment Analysis in Adversarial Games
Fuzzy Search Strategy Generation for Adversarial Systems Using Fuzzy Process Particle Swarm Optimization, Fuzzy Patterns, and a Hunch Factor
Fuzzification of the Special Simplex Method for the Transportation Problem
Fuzzification of the Golden Ratio Search (and preliminary research on other search algorithms)
Fuzzification of Simple Sorting AlgorithmsSlide31
Results and Key Observations
A fuzzy algorithm:Often performs better than the traditional algorithm because it takes advantage of the quickness of integer calculations over double-precision calculations
Finds multiple
“
good enough
”
solutions within a controllable range faster than the traditional algorithms
Is useful when the end user would prefer several similar answers to select from and not just one answerSlide32
Future Work: Questions to be Answered
Continue exploration into the use of Fuzzy AlgorithmsWhat algorithms benefit from being made fuzzy and how can the gains be measured?
What algorithm characteristics make it a good candidate for becoming a fuzzy algorithm?
How do these characteristics impact the solution of a problem when a fuzzy algorithm is used to solve it?
Refining and further investigation into the Hunch Factor
What membership functions work best? In what situations does it work best? What manipulation works best? etc.Slide33
Picture Citations (in order of appearance)
https://www.technologyreview.com/s/601519/how-to-create-a-malevolent-artificial-intelligence/https://fahmirahman.wordpress.com/2011/04/19/the-history-of-the-eniac-computer/
https://en.wikipedia.org/wiki/ENIAC
http://themadraspost.blogspot.com/2014/02/unknown-electronic-do-you-know-which-is.html
http://www.computerhistory.org/revolution/birth-of-the-computer/4/78
http://www.columbia.edu/cu/computinghistory/eniac.html
http://www.phillyvoice.com/70-years-ago-six-philly-women-eniac-digital-computer-programmers/
http://www.fanpop.com/clubs/star-trek-the-next-generation/images/31158790/title/data-wallpaper
http://www.dailymail.co.uk/tvshowbiz/article-2762469/Anthony-Daniels-reveals-initially-refused-play-C-3PO-voice-role-new-Star-Wars-movie.html
https://s-media-cache-ak0.pinimg.com/originals/28/91/aa/2891aa51b8fbb124a1478ae03a8008aa.jpg
http://coralis101.deviantart.com/art/Wall-E-and-EVE-Icons-97034881
http://also.kottke.org/misc/images/super-toy-teddy.jpg
https://www.cloudynights.com/uploads/profile/photo-224707.jpg?_r=0
https://www.mathworks.com/company/newsletters/articles/alan-turing-and-his-connections-to-matlab.htmlhttp://www.impawards.com/2014/imitation_game_ver6.htmlhttp://zadeh.narod.ru/Lotfi_Zadeh_L.jpg