Martha Larson: AI vs. Our Eye: New Perspectives on Adversarial Images – 10 March 13.15–14.30

This I2S seminar series lecture explores different approaches to answering the questions: Does AI really see images the way that people do? What are the consequences of the difference between AI and our eye?

Abstract

What do we see when we look at an image? It depends on the context and on our point of view. Today’s Artificial Intelligence (AI) technology astounds us with its ability to automatically interpret the contents of images. But does AI really see images the way that people do? What are the consequences of the difference between AI and our eye? This talk explores different approaches to answering these questions, putting a special focus on computer vision used to interpret the type of images people share online. We look at some of the shortcomings of current AI technology in understanding images. We also look at adversarial images that are easily interpreted by humans, but posed difficulties for computers, and discuss how they might actually be useful for protecting privacy. Finally, we discuss the MediaEval Benchmarking Initiative for Multimedia Evaluation in which researchers are attempting to make progress on some of the interesting open questions of AI for image interpretation.

Biography

Martha Larson works in the area of multimedia retrieval and recommender systems. She is a professor at Radboud University in the area of Multimedia Information Technology and is also affiliated with the Multimedia Computing Group at Delft University of Technology. Previously, she researched and lectured in the area of audio-visual retrieval at Fraunhofer IAIS and at the University of Amsterdam. Larson is co-founder of the MediaEval Benchmarking Initiative for Multimedia Evaluation.

Image: Martha Larson