Machine Learning Damon Waring 22 April 2003. 2 of 15 Agenda Problem, Solution, Benefits Problem,...
-
Upload
edmund-harrell -
Category
Documents
-
view
240 -
download
0
Transcript of Machine Learning Damon Waring 22 April 2003. 2 of 15 Agenda Problem, Solution, Benefits Problem,...
Machine Machine LearningLearning
Damon WaringDamon Waring
22 April 200322 April 2003
2 of 15
AgendaAgenda
Problem, Solution, BenefitsProblem, Solution, Benefits Machine Learning Overview/BasicsMachine Learning Overview/Basics Face detection, recognition, and Face detection, recognition, and
demodemo How this applies to usHow this applies to us SummarySummary
3 of 15
ProblemProblem
Software frequently requires users Software frequently requires users or developers to do simple, or developers to do simple,
repetitive tasksrepetitive tasks
4 of 15
SolutionSolution
Machine LearningMachine Learning ““The study of computer algorithms The study of computer algorithms
that improve automatically through that improve automatically through experience” –Tom Mitchell, experience” –Tom Mitchell, Machine Machine LearningLearning
Machine learning uses include: Machine learning uses include: Security (Pattern recognition, face Security (Pattern recognition, face
recognition)recognition) Business (Stocks, user behaviors)Business (Stocks, user behaviors) Medical (Research)Medical (Research) Ease of Use (Focus of this presentation)Ease of Use (Focus of this presentation)
Algorithms that execute based on experienceAlgorithms that execute based on experienceAlgorithms that execute based on experienceAlgorithms that execute based on experience
5 of 15
BenefitsBenefits
Makes human-computer interaction Makes human-computer interaction easiereasier
Relatively simple to integrateRelatively simple to integrate Will distinguish your product from Will distinguish your product from
othersothers Increase customer satisfactionIncrease customer satisfaction Will improve simple intelligent Will improve simple intelligent
systems (ex: Microsoft Word’s systems (ex: Microsoft Word’s grammar checker)grammar checker)
Enhances the user experienceEnhances the user experienceEnhances the user experienceEnhances the user experience
6 of 15
High Level Operation:High Level Operation:Recognition AlgorithmsRecognition Algorithms
Training ModeTraining Mode Training SetTraining Set Iteratively analyze Iteratively analyze
inputs and refine inputs and refine algorithmalgorithm
Store learned data Store learned data
Operation ModeOperation Mode New inputNew input Process input using Process input using
learned datalearned data Produce a decisionProduce a decision
Recognition algorithms are taught and react like humansRecognition algorithms are taught and react like humansRecognition algorithms are taught and react like humansRecognition algorithms are taught and react like humans
““Learn from nature. It has had 4 billion years Learn from nature. It has had 4 billion years to develop its techniques” – My Dadto develop its techniques” – My Dad
7 of 15
Case Study: Artificial Case Study: Artificial Neural NetworkNeural Network
Takes N inputsTakes N inputs Calculates the Calculates the
weight each input weight each input has on final decisionhas on final decision
Neuron outputs a 1 Neuron outputs a 1 if the decision is if the decision is true, 0 if it is falsetrue, 0 if it is false
Groups of neurons Groups of neurons make up an make up an artificial neural artificial neural networknetwork
Group of weighted input values determine a binary outputGroup of weighted input values determine a binary outputGroup of weighted input values determine a binary outputGroup of weighted input values determine a binary output
8 of 15
Face DetectionFace Detection
1.1. Image pyramid used to locate faces of different sizesImage pyramid used to locate faces of different sizes2.2. Image lighting compensationImage lighting compensation3.3. Neural Network detects rotation of face candidateNeural Network detects rotation of face candidate4.4. Final face candidate de-rotated ready for detectionFinal face candidate de-rotated ready for detection
9 of 15
Face Detection (Con’t)Face Detection (Con’t)
5.5. Submit image to Neural NetworkSubmit image to Neural Networka.a. Break image into segmentsBreak image into segmentsb.b. Each segment is a unique input to the networkEach segment is a unique input to the networkc.c. Each segment looks for certain patterns (eyes, Each segment looks for certain patterns (eyes,
mouth, etc)mouth, etc)
6.6. Output is likelihood of a faceOutput is likelihood of a face
10 of 15
Face Recognition and Face Recognition and demodemo
Demo: Hidden Markov Model Face Demo: Hidden Markov Model Face RecognitionRecognition Observes location of facial features with Observes location of facial features with
respect to each otherrespect to each other Person is found through unique Person is found through unique
“fingerprint” created by distances “fingerprint” created by distances between featuresbetween features
Demo is from OpenCV – Intel’s open Demo is from OpenCV – Intel’s open source computer vision librarysource computer vision library
Implementations vary widely and have different success ratesImplementations vary widely and have different success ratesImplementations vary widely and have different success ratesImplementations vary widely and have different success rates
11 of 15
Adobe Photoshop AlbumAdobe Photoshop Album
Software that organizes digital Software that organizes digital picturespictures
Tags are dragged to each photo to Tags are dragged to each photo to categorize itcategorize it
Tagging 100’s of photos is tediousTagging 100’s of photos is tedious Face recognition could automatically Face recognition could automatically
tag photos or replace tags altogethertag photos or replace tags altogether
Machine learning can be used to make everyday apps easierMachine learning can be used to make everyday apps easierMachine learning can be used to make everyday apps easierMachine learning can be used to make everyday apps easier
12 of 15
Current Uses of MLCurrent Uses of ML
DivX – Face detectionDivX – Face detection POV-Ray – Neural Net learns memory POV-Ray – Neural Net learns memory
accessesaccesses Ancestry.com – Uses Optical Ancestry.com – Uses Optical
Character Recognition to digitize Character Recognition to digitize newspapersnewspapers
Deep Blue Junior – Less powerful than Deep Blue Junior – Less powerful than Deep Blue, but smarter because of Deep Blue, but smarter because of Neural NetworksNeural Networks
13 of 15
Other AreasOther Areas
Artificial Intelligence (AI)Artificial Intelligence (AI) Data MiningData Mining Fuzzy LogicFuzzy Logic Optical Character Recognition (OCR)Optical Character Recognition (OCR)
14 of 15
SummarySummary
Machine learning is possible todayMachine learning is possible today Large amounts of research are Large amounts of research are
availableavailable Quality open source code available in Quality open source code available in
some areassome areas Will require time and creativity to Will require time and creativity to
implementimplement Why do it? Makes human-computer Why do it? Makes human-computer
interface simplerinterface simpler
15 of 15
ReferencesReferences BooksBooks
Machine LearningMachine Learning by Tom Mitchell ( by Tom Mitchell (http://www-2.cs.cmu.edu/~tom/mlbook.htmlhttp://www-2.cs.cmu.edu/~tom/mlbook.html )) Web sitesWeb sites
Hidden Markov Models Hidden Markov Models http://jedlik.phy.bme.hu/~gerjanos/HMM/node2.htmlhttp://jedlik.phy.bme.hu/~gerjanos/HMM/node2.html Links recommended by PCAI Links recommended by PCAI http://www.ics.uci.edu/~mlearn/MLOther.htmlhttp://www.ics.uci.edu/~mlearn/MLOther.html CMU’s research areas (scroll down): http://www.ri.cmu.edu/people/kanade_takeo.htmlCMU’s research areas (scroll down): http://www.ri.cmu.edu/people/kanade_takeo.html MIT’s Media Lab: http://www.media.mit.edu/MIT’s Media Lab: http://www.media.mit.edu/ Computer vision links: http://www-2.cs.cmu.edu/afs/cs/project/cil/ftp/html/vision.htmlComputer vision links: http://www-2.cs.cmu.edu/afs/cs/project/cil/ftp/html/vision.html Open source computer vision library (OpenCV): Open source computer vision library (OpenCV):
http://sourceforge.net/projects/opencvlibrary/http://sourceforge.net/projects/opencvlibrary/ JournalsJournals
PCAI (a great industry magazine, web site is bad)- http://www.pcai.comPCAI (a great industry magazine, web site is bad)- http://www.pcai.com ScienceDirect (http://www.sciencedirect.com) “Computer Vision and Image ScienceDirect (http://www.sciencedirect.com) “Computer Vision and Image
Understanding,” “Artificial Intelligence,” “Neural Networks”Understanding,” “Artificial Intelligence,” “Neural Networks” IEEE Proceedings (http://www.ieee.org) “Pattern Analysis and Machine Intelligence,” IEEE Proceedings (http://www.ieee.org) “Pattern Analysis and Machine Intelligence,”
“Image Processing”“Image Processing” IEEE Papers/Proceedings referenced in this presentationIEEE Papers/Proceedings referenced in this presentation
Hidden Markov Models (used in OpenCV Demo) “Maximum likelihood training of the Hidden Markov Models (used in OpenCV Demo) “Maximum likelihood training of the embedded HMM for face detection and recognition.” Nefian, A.V.; Hayes, M.H. III; embedded HMM for face detection and recognition.” Nefian, A.V.; Hayes, M.H. III; Image Processing, 2000. Proceedings. 2000 International Conference on, Volume: 1, Image Processing, 2000. Proceedings. 2000 International Conference on, Volume: 1, Pages 33-36.Pages 33-36.
““Neural network-based face detection.” Rowley, H.A.; Baluja, S.; Kanade, T; Pattern Neural network-based face detection.” Rowley, H.A.; Baluja, S.; Kanade, T; Pattern Analysis and Machine Intelligence, IEEE Transactions on, Volume 20 Issue 1, Jan Analysis and Machine Intelligence, IEEE Transactions on, Volume 20 Issue 1, Jan 1998. Pages 23-38. (Paper posted at: 1998. Pages 23-38. (Paper posted at: http://www.ri.cmu.edu/projects/project_271.htmlhttp://www.ri.cmu.edu/projects/project_271.html ))
““Rotation Invariant Neural Network-Based Face Detection” Rotation Invariant Neural Network-Based Face Detection” http://www.ri.cmu.edu/projects/project_271.htmlhttp://www.ri.cmu.edu/projects/project_271.html