Machine Learning meets the Real World:Successes and new research directions
Data, data everywhere...
A wealth of information
Machine learning success(Machine learning is ubiquitous)
Why research in machine learning is so good today
Plan for this talk
Induction of decision trees
Inductive learning
Sample data
Predictive modelI.e., g<x>
Learning objectives
TDIDT
Which is better?
The Gain Criterion
Information (Entropy)
Entropy after a split
Information Gain
Scrubber (the success story)
MAX, 1990
Scrubber 2
Scrubber 3
Implementation difficulties
Requirements
Additional requirements (ours)
Phase I: Modeling Scrubber 2
Data
Background knowledge
Phase I results
Phase II: Acceptance
Trading off simplicity and correctness
Phase II results
Phase III: Working toward extensibility
Phase IIIb: More data
Phase III results
Summarizing the success story
Lessons can be learned from success
Further reading and acknowledgements
Email: andrea@cs.williams.edu
Home Page: http://www.cs.williams.edu/~andrea
Download presentation source