This week's Statistics Seminar speaker will be Eric Bax from Yahoo, Inc.
Talk Title: Validation of k-NN Classifiers and Pricing Information
Theory: Validation of k-nearest neighbor classifiers
Recent error bounds for k-nn classifiers begin with a holdout bound and then use inclusion and exclusion to estimate the difference in error rates between the holdout classifier and the classifier that uses all in-sample data. Using this approach, we get O(1/sqrt(n) (ln n) (ln n)) difference between in-sample and out-of-sample error rates, similar to bounds for other classifiers such as artificial neural networks and support vector machines.
Applied: Pricing Information
In online advertising marketplaces, third party data providers supply information to advertisers about which users are likely to respond to their ads. So advertisers buy a combination of media from publishers and information from third parties. How should it all be priced? We will have a hands-on exercise to explore this question.