Sparse autoencoders reveal selective remapping of visual concepts during adaptation

Open in new window