Anomaly detection and simulation for unlabeled sensor data

The rapid development in the areas of statistics and machine learning demonstrate unprecedented performance in making cognitive business decisions. Quartic.ai aims to use state-of-the-art machine learning technology to help manufacturers assess and maintain the quality of their industrial units, which suffer damage due to continuous usage and normal wear and tear. Such damage needs to be detected early to prevent further losses. The data in this domain are recorded using sensors at various stages in the process flow. Major challenges of analyzing these sensor data are (1) unlabeled data, which may contain very few unobserved anomalies or outliers; (2) the development and evaluation of algorithms that can robustly detect anomalies. Due to the lack of labels, the performance of algorithms can not be directly evaluated. To tackle the problems, we will carefully design simulations by taking into account of various types of outliers and develop novel robust one-class classification algorithms.

Intern: 
Sile Tao
Faculty Supervisor: 
Linglong Kong
Province: 
Alberta
Partner University: 
Discipline: 
Program: