Adrien Doerig

The paper identifies auditory representations during silent reading using EEG and computational modeling, showing they emerge before word onset and are stronger for unexpected words.

The paper shows that Capsule Neural Networks (CapsNets) with recurrent grouping and segmentation outperform feedforward and standard recurrent CNNs in modeling human global shape processing using visual crowding.

This study uses visual crowding to show that feedforward CNNs lack human-like global shape processing due to architectural limitations.