Adversarial Machine Learning: When Good Learning Leads to Bad Security

  • Duthie Center Room 117

With the advancement of machine learning and its promise to provide good generalization capability, its usage has become ubiquitous. The over-enthusiasm to develop and deploy a data-based solution has led to the overlooking of the flaws of machine learning itself.

Machine learning models operating in a non-stationary and adversarial environment, is susceptible to evasion attacks at test time. In this talk we will discuss our preliminary results on the vulnerability of classifiers to evasion attacks, even when the only access to the system available is a black box interface. 


Tegjyot Singh Sethi is a doctoral candidate in the Data Mining Lab, at the University of Louisville. He completed his undergraduate studies in India at GITAM University in 2012, and his MS in Computer Science at U of L in 2013. His research interests include: Streaming data, concept drift detection, adversarial machine learning, bio inspired algorithm and scalable big data systems. 

Not familiar with the University of Louisville campus? Here's a Campus Map to help you find your way.

This talk is sponsored by the University of Louisville Department of Computer Engineering and Computer Science in cooperation with IEEE Computer Society Louisville Chapter.