Vasilis A. Sotiris

For my thesis I studied and modeling latent mechanisms/phenomena that drive an observable process until some endpoint of interest. In survival and reliability analysis such endpoints are typically the time of failure. The data driven statistical nature of these models make them generally applicable across many areas, including finance, economics, medicine, etc.

Stochastic processes are the theoretical foundation for modeling the latent (unobserved) event-causing mechanisms. Parametric and predictive inference equations are developed based on the first hitting times of the stochastic processes to a random/uncertain barrier set (threshold). In addition, modeling also accommodates time dependent covariates, which vary with time, environment, usage, etc., and are used to improve inferences and predictions.

Click here to view a sample matlab prediction plot. The asterisks represent a latent process, and the density curves the conditionally predicticted distribution for the degradation variable.

Previously, I worked with machine learning algorithms for anomaly detection and prediction. I developed a one-class Bayesian support vector machine algorithm to detect the onset of system anomalies,and trend output classification probabilities, as a way to monitor the health of a system. Click here to see a snap-shot of the Bayesian support vector machine matlab tool. This work involved theory on support vector machines, Bayesian linear models, and gaussian processes.

V. Sotiris, E. Slud, M.Pecht, "Reliability Inference in a First Hitting Time Model with Augmented Data", Mathematical Methods in Reliability, June 2011 abstract

V. Sotiris, T. Chow, M. Pecht, "Expected Time to Failure Estimation in Laptop Computers Exhibiting Intermittent Errors", IEEE Transactions on Reliability, 2010 (In Review) abstract

V. Sotiris, P. Tse, M. Pecht "Anomaly Detection Through a Bayesian Support Vector Machine", IEEE Transactions on Reliability, Vol. 59, No. 2, 2010
abstract

V. Sotiris, M. Pecht "Anomaly Prediction Through Latent Variable Gaussian Process Models with SVM Mean Function Priors", International Workshop on Structural Health Monitoring, Stanford University, Palo Alto, CA. Sept 2009

S. Kumar, V. Sotiris, and M. Pecht, "Health Assessment of Electronic Products using Mahalanobis Distance and Projection Pursuit Analysis" International Journal of Computer, Information, and Systems Science, and Engineering, Vol. 2, No. 4, pp. 242-250, Fall 2008

V. Sotiris, M. Pecht, "Support Vector Prognostics Analysis of Electronic Products and Systems", Association for the Advancement of Artificial Intelligence, pp. 120-127, 2007