Machine Learning Meetup Notes: 2010-05-26

• Andy gave overview of where we're at with KDD data
• Mike S gave presentation:
  • Gaussian Mixture Models
  • k-means clustering
  • very basic expectation-maximization
• Brainstorming session on how to reduce skill set column
  • Tom tried to quantify opportunity per skills per row as high dimensional vector
• Brainstorming on how to reduce other data and compute new features for the KDD Dataset
  • Tom will apply k-means clustering of skills (or steps), for data reduction
  • Andy will compute new features: unique step/problem id, student IQ (avg. correct), step challenge/difficulty (avg correct), step complexity (# skills required)
  • Mike will use self-organizing maps to reduce skills
  • Paul will visualize/summarize the data, to provide understanding and insight
  • Mike will set up an FTP server for people to transfer their enormous datasets
  • Theo will use some Weka classifiers to produce a classification method for the data