»ÆÉ«¶ÌÊÓƵ

prof. Jeffrey Bowers

Deep Problems with Neural Network Models of Human Vision and Language

Abstract: Deep neural networks (DNNs) developed in computer science are successful in a range of vision and language tasks and can predict brain activations of humans (and macaques) better than alternative models. This has led to the common claim that DNNs are the best models of biological vision and langauge. Here I show that the success of these models in predicting brain activations is a poor metric for judging the similarity of DNNs and brains, and indeed, these models account for few findings in psychology in the domain vision, and they show similar problems when it comes to language.