The title of this team's award-winning paper is "Distorted Representation Space Characterization through Backpropagated Gradients,” and it was chosen from a field of more than 900 papers for the 2019 IEEE ICIP Best Paper Award. Kwon and Prabhushankar are ECE Ph.D. students, and Temel is an ECE postdoctoral fellow. They all work in the Omni Lab for Intelligent Visual Engineering and Science (OLIVES), which is led by ECE Professor Ghassan Al-Regib.  

In recent years, artificial intelligence (AI) has shown to be effective in a number of applications. Throughout all of this excitement, the focus has been on the information that is learned by the AI systems. In reality, knowing the information that was not learned is equally important. 

The OLIVES team's work in this paper characterizes the information that has not been learned by an AI system. This characterization is particularly important to understand the tasks that an AI system cannot successfully perform, which paves the way to AI interpretability. 

AI interpretability is crucial for its deployment in day-to-day activities and the realization of the 4^th industrial revolution. This work can also be applied to diverse visual applications, which require robust AI systems. These applications include, but are not limited to, medical image diagnosis, surveillance, autonomous vehicles, texture analysis, seismic interpretation, image/video classification, and object detection.

Photo cutlines (photos provided by Ghassan AlRegib)

Top photo: ECE Ph.D. student Mohit Prabhushankar presents his team’s work at the 2019 IEEE ICIP.

Middle photo: ECE Ph.D. student Gukyeong Kwon presents his team’s work at the 2019 IEEE ICIP.

Bottom photo: Members of the OLIVES Lab attended the 2019 IEEE ICIP. Pictured left to right are ECE Postdoctoral Fellow Can Temel, Charlie Lehman, Gukyeong Kwon, Jinsol Lee, Mohit Prabhushankar, and ECE Professor Ghassan AlRegib. Lehman, Kwon, Lee, and Prabhushankar are all ECE Ph.D. students.