We were pleased to welcome Prof. Dr. Felix Wichmann from University of Tübingen to our scientific seminar series, presenting on whether modern DNNs are adequate models of human object recognition.

Current DNNs match human-level performance on large-scale image classification, and a growing body of research indicates shared processing signatures between artificial and biological vision systems. The Wichmann Lab put this claim to the test and showed: Adversarial susceptibility, texture bias, error consistency differences, and context bias point to significant differences in object recognition between humans and DNNs.

In his compelling and entertaining talk, Felix concluded: DNNs are highly valuable scientific tools, but not adequate models of human visual perception, at least not for object recognition.

We thank Felix for this grounded, data-driven perspective on a question that shapes how we think about both artificial and biological vision, and HRI-EU associate Christiane Wiebel-Herboth for establishing the contact!