Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session C: Friday, August 15, 2:00 – 5:00 pm, de Brug & E‑Hall

Replicating Posner & Keele (1968) with Standard Convolutional Neural Networks

Milan Van Maldegem¹, Gido M van de Ven², Hans Op de Beeck¹, Tinne Tuytelaars¹; ¹KU Leuven, ²Delft University of Technology

Presenter: Milan Van Maldegem

This study tests whether standard convolutional neural networks (CNNs) replicate the human prototype effect in category learning. Using a setup based on Experiment III of Posner & Keele (1968), three CNN architectures were trained on abstract dot patterns. While none matched the human results, AlexNet and DenseNet-121 showed better accuracy for unseen prototypes than for new exemplars, suggesting a weak prototype bias. These results provide a foundation for further research on category learning in humans and CNNs.

Topic Area: Visual Processing & Computational Vision

Extended Abstract: Full Text PDF