1 / 25

Pattern Recognition Final Task

Pattern Recognition Final Task. Ibrahim Arief – 185099 Timo Eckhard – 185126 University of Joensuu December 17 th , 2009. Contents. M-Fold-Cross Training Color Data Preprocessing Bayesian Classifier Multilayer Perceptron K-Means Clustering Speech Data Preprocessing

hagop
Download Presentation

Pattern Recognition Final Task

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. Pattern RecognitionFinal Task Ibrahim Arief – 185099 Timo Eckhard – 185126 University of Joensuu December 17th, 2009

  2. Contents • M-Fold-Cross Training • Color Data • Preprocessing • Bayesian Classifier • Multilayer Perceptron • K-Means Clustering • Speech Data • Preprocessing • Bayesian Classifier • Multilayer Perceptron • K-Means Clustering • Summary

  3. M-Fold-Cross Training • Partition into M subsets • One subset is assigned as test subset, the rest is training subset • We use the training subset for testing against test subset • Assign other subset as new test subset, the rest is training subset for that particular one • Repeat until all partition took their turn being tested

  4. Spectral Color Data - Preprocessing

  5. Spectral Color Data – Bayesian Classifier (1) • Raw spectral input – all classified to class 3

  6. Spectral Color Data – Bayesian Classifier (2) • Preprocessing : Tristimulus • Nice clumping, linearly separable

  7. Spectral Color Data – Bayesian Classifier (3) • Very high accuracy : 99.97%

  8. Spectral Color Data – Multi Layer Perceptron • Raw spectral data as input : ~5% • Tristimulus as input : ~30% • Question : parameters? • Answer : exhaustive search?

  9. Spectral Color Data – K-Means Clustering

  10. Speech Data – Preprocessing (1) • MFCC – Timeseries? • Plot of coefficients within a class

  11. Speech Data – Preprocessing (2) • Plot of variance for each coefficient

  12. Speech Data – Preprocessing (3) • Plot of bayesian accuracy for n-least-varied

  13. Speech Data – Preprocessing (4) • Delta-coefficients • Source: http://cslu.cse.ogi.edu/fsj/issues/issue5/sparse-ann/PhoneProbEst.html • Formula • Dimensionality reduction

  14. Speech Data – Bayesian Classifier • Frequency matters • No risk matrix • Raw accuracy : 18.13% • Delta-coefficient preprocessing : 96.06%

  15. Speech Data – Multi Layer Perceptron • Hidden Neuron : 22 • Normalized Raw Data : 20.25% • Reduced dimension, delta coefficient : 29.52% • Delta coefficient without reduced dimension : 27.84%

  16. Summary – Spectral Color Data • Bayesian Classifier • Raw Data : 3.92% • Preprocessed : 99.97% (tristimulus) • Multi Layer Perceptron • Raw Data : ~5% • Preprocessed : 58.1% (tristimulus) 99.7% (tristimulus + CIELAB + sRGB) • K-Means Clustering • Raw data : 92% • Preprocessed : 95%

  17. Summary – Speech Data • Bayesian Classifier • Raw Data : 18.19% • Preprocessed : 96.09% (delta-derivative, high variance elimination) • Multi Layer Perceptron • Raw Data : 20.25% • Preprocessed : 29.52% (delta-derivative, high variance elimination) • K-Means Clustering • Raw data : 24% • Preprocessed : 62% (normalized, delta-derivative)

More Related