Login
Talks & Seminars
Title: Generative Models for Frequent Pattern Mining
Dr. Srivatsan Laxman, Microsoft Research Labs, Bangalore
Date & Time: November 18, 2008 11:30
Venue: Conference Room, ā€˜Cā€™ Block, 01st floor, Kanwal Rekhi Building
Abstract:
Patterns are local structures in data that capture information about a small number of attributes or data points. Pattern discovery (or unearthing of frequent patterns in data) is a popular task of data mining. In this talk, we will explore a role for generative models in frequent pattern mining. We begin with the temporal data mining framework of frequent episode discovery. By introducing the notion of non-overlapped occurrences-based frequency for episodes, it is possible to formally connect frequent (serial) episode discovery with HMM learning. The non-overlapped occurrences-based frequency also allows for very efficient algorithms for frequent episode discovery. Further, it leads to a statistical significance test for episodes based on episode frequency. The thresholds derived from the significance test were found to be effective for root-cause analysis of faults logged in automotive engine assembly lines of General Motors. Similar analysis has been developed for the frameworks of frequent generalized episodes (where event durations are incorporated in the episode structure) and frequent itemsets (in transaction databases) as well. The most common use of frequent patterns in data mining is to derive rules (such as association rules) that hold in the data. Statistical interpretations and generative models for pattern mining, can lead to wider application of pattern discovery in clustering, prediction, etc. For example, it is possible to build generative models for sequence prediction based on a mixture of HMMs (with each frequent serial episode associated with one component of the HMM mixture). This approach to sequence prediction was successfully applied for predicting targeted user-behavior on the web.
Speaker Profile:
Srivatsan Laxman received the PhD degree in electrical engineering from the Indian Institute of Science, Bangalore, in 2006. He is currently an Associate Researcher in the CSA Group (Cryptography, Security and Applied Mathematics) at Microsoft Research Labs, Bangalore. His research interests include data mining, machine learning, pattern recognition, and signal processing. Details about his research interests as well as selected publications can be found at http://research.microsoft.com/~slaxman.
List of Talks

Webmail

Username:
Password:
Faculty CSE IT
Forgot Password
    [+] Sitemap     Feedback