Oregon State University



Event Details

MS Final Exam, Non Thesis, Tadesse Zemicheal

Tuesday, March 21, 2017 10:00 AM - 12:00 PM

Sensor-dx: A framework for multi-view based anomaly detection and diagnosis of sensors in weather stations
In this project, we develop an end-to-end modular machine learning framework for automated quality control in a weather station. A multi-view data is constructed from a stream of time-series weather data, with each view tuple generated from a set of sensor-state-variable. An anomaly detection instance (ADI) is used to apply time-series decomposition, and generate anomaly score of constructed views. A non-parametric kernel density estimator (KDE) probability model is trained to fit training data of sensor state configuration. Finally, a probabilistic inference of the latent sensor state variable and diagnosis is done using a greedy algorithm. Our initial result shows, the number of false alarm are smaller using multi-view compared to single view anomaly detection.

Major Advisor: Thomas Dietterich
Committee: Alan Fern
Committee: Raviv Raich

Kelley Engineering Center (campus map)
Calvin Hughes
1 541 737 3617
Calvin.Hughes at oregonstate.edu
Sch Elect Engr/Comp Sci
This event appears on the following calendars: