Human alignment of neural network representations

Lukas Muttenthaler, Jonas Dippel, Lorenz Linhardt +2

Today's computer vision models achieve human or near-human level performance across a wide variety of vision tasks. However, their architectures, data, and learning algorithms diff…