PDMC Entry

1
PDMC Entry

• By
• Saurabh Amin
• Dept. of Civil Eng., UT Austin
• Presenter
• Nicholas K. Jong
• Dept. of Computer Sciences, UT Austin

With help of Gunjan Gupta, Dept. of Elect.
Comp. Eng. UT Austin
2
What was planned

3
What was done

• Bayesian Prior Model for context identification
• Preprocessing
• Remove all sessions that have no labeled
  sub-sessions
• Split 70 for training and 30 for testing
• Select labeled sub-sessions
• Assign class labels Context1, Context2 , Others
• Model Building
• Count (Characteristic 1) for each class label
• Fit 3-parameter Weibulls. Size and scale
  parameters fixed by observation and shape
  parameter (the most critical one) fixed by
  observing performance on testing set
• Apply Bayes theorem Pr(ConChar)P(CharCon)P(Co
  n)/P(Char)
• Performance Evaluation
• Evaluate performance on scoring criterion (fixed
  by organizers)
• Compare with the most naïve global prior model
  Better performance
• Similar procedure repeated for gender prediction

4
What is to be done

• Feature extraction and classification
• Aggregated log-transformed 32-point FFT plot