Presenter: Cynthia Rudin 

Abstract: I will describe two methods and applications for pattern detection, where patterns are grown from a seed of a few items:

1) Growing a List: The next generation of search engines should not simply retrieve URLs, but should aim at retrieving information. We designed a system that leads into this next generation, leveraging information from across the Internet to grow an authoritative list on almost any topic, starting from a seed.

2) Crime Series Detection: In joint work with the Cambridge Police Department, we designed a method called "Series Finder" that detects patterns of crime that are committed by the same individual or group. The method is tested on a decade's worth of housebreak data from Cambridge, MA.

Series Finder is a supervised pattern detection algorithm. Time permitting, I will provide statistical learning theoretic guarantees for supervised pattern detection methods.

Collaborators are: Benjamin Letham, Katherine Heller, Tong Wang, Daniel Wager, Rich Sevieri, and Jonathan Huggins

References are here:
-----
Growing a List
Paper: http://web.mit.edu/rudin/www/LethamRuHe13.pdf

Boston public radio interview about the project "A New Way to Google": 
http://www.wgbhnews.org/post/new-way-google

Learning to Detect Patterns of Crime
Paper: http://web.mit.edu/rudin/www/WangRuWaSeECML13.pdf

Boston Globe article "Cambridge Police Look at Math to Solve Crimes":
http://www.bostonglobe.com/metro/2013/08/03/cambridge-police-look-math-s...

Towards a Theory of Pattern Discovery
Paper: http://web.mit.edu/rudin/www/HugginsRu2013.pdf