The OSU Computer Science and Engineering Department

Home | Events | People

	Department
	Calendar
		Calendar Index
		Upcoming Speakers

	News
	Research
	Undergrad Programs
	Graduate Programs
	People
	Courses
	Computing Services

Systems Presentation

Double Header!

LOADED: LINK-BASED OUTLIER & ANOMALY DETECTION IN EVOLVING DATA SETS	INTERNSHIP EXPERIENCE
Amol Ghoting OSU-CSE	Weikuan Yu OSU-CSE

Friday, Oct. 15th
3:30pm, 480 Dreese Labs
All interested parties are invited.
Pizza will be served after the talk.

Abstract:
Abstract: Detecting outliers or anomalies efficiently is an important problem in many areas of science, medicine and information technology. Applications range from data cleaning to fraud and intrusion detection. Most approaches to date have focused on detecting outliers in a continuous attribute space. However, almost all real-world data sets contain a mixture of continuous and categorical attributes, and the categorical attributes are either ignored or handled in an ad-hoc manner by current approaches, resulting in a loss of information. The challenge is to efficiently identify outliers in mixed attribute data sets under a variety of constraints, such as minimizing the time to respond, and adapting to the data influx rate. To address this challenge, we present LOADED, a one-pass algorithm for outlier detection in evolving data sets containing both continuous and categorical attributes. LOADED is a tunable algorithm, wherein one can trade off computation for accuracy so that domain-specific (e.g. intrusion detection) response times are achieved. Experimental results validate the effectiveness of our schemes over several real data sets. LOADED provides very good detection and false positive rates, which are several times better than those provided by existing distance-based schemes. We are currently extending LOADED to efficiently find outliers in a distributed setting.

Abstract: Open MPI is a project recently initiated to provide a fault-tolerant, multi-network capable, and production-quality implementation of MPI-2 interface based on experiences gained from FT-MPI, LA-MPI, LAM/MPI, and MVAPICH projects. Its initial communication architecture is layered on top of TCP/IP. In this project, we have designed and implemented Open MPI point-to-point layer on top of a high-end interconnect, Quadrics/Elan4. Design challenges related to dynamic process/connection management, utilizing Quadrics RDMA capabilities and supporting asynchronous communication progression are overcome with salient strategies to utilize Quadrics Queued-based Direct Memory Access (QDMA) and Remote Direct Memory Access (RDMA) operations, along with the chained event mechanism.

Experimental results indicate that the resulting point-to-point transport layer implementation is able to achieve comparable performance to Quadrics native QDMA operations, from which it is derived. While not taking advantages of Quadrics/Elan4 NIC-based tag matching due to its design requirements, this point-to-point transport layer provides a high performance implementation of MPI-2 compliant message passing over Quadrics/Elan4.

These talks are funded by a grant from the Ohio Board of Regents.