Ivans Lubenko

Doctoral Student

Student, Wadham College

Wolfson Building, Parks Road, Oxford OX1 3QD

Interests

My research revolves around studying the practical limits of machine learning applied to unusual problems such as intrusion detection or steganalysis. My DPhil project began with exploring the application of Logistic Regression to steganalysis. I am intending to find whether the obvious advantages of Logistic Regression, the output probability estimates, make it a better fit to steganalysis than the more often used SVMs. The common steganalysis requirement of estimating confidence in detection can, for example, be satisfied by using the Logistic Regression probability estimates.

Another direction of my research is understanding the modern steganalytic features space. I began with exploring its properties using graphical representation and discovered a surprising amount of collinearity between them. This gives a reason to believe that there is a significant geometric aspect to steganalytic features, which can be exploited to build better detectors. I am planning to experiment with modifying, say, Logistic Regression to take into account these properties of the feature space with the aim of building a more general steganalytic detector.