It takes neurons to understand neurons: Digital twins of visual cortex synthesize neural metamers
Cobos, E.; Muhammad, T.; Fahey, P. G.; Ding, Z.; Ding, Z.; Reimer, J.; Sinz, F. H.; Tolias, A.
Show abstract
Metamers, images that are perceived as equal, are a useful tool to study representations of natural images in biological and artificial vision systems. We synthesized metamers for the mouse visual system by inverting a deep encoding model to find an image that matched the observed neural activity to the original presented image. When testing the resulting images in physiological experiments we found that they most closely reproduced the neural activity of the original image when compared to other decoding methods, even when tested in a different animal whose neural activity was not used to produce the metamer. This demonstrates that deep encoding models do capture general characteristic properties of biological visual systems and can be used to define a meaningful perceptual loss for the visual system.
Matching journals
The top 3 journals account for 50% of the predicted probability mass.
Similar papers in this journal
Similar papers in this journal
Similar papers in this journal
- An unsupervised map of excitatory neurons' dendritic morphology in the mouse visual cortex 96%
- Large-scale capture of hidden fluorescent labels for training generalizable markerless motion capture models 95%
- Intrinsic dynamics enhance temporal stability of stimulus representation along rodent visual cortical hierarchies 95%
Similar papers in this journal
- Emergence of brain-like mirror-symmetric viewpoint tuning in convolutional neural networks 96%
- CEM500K - A large-scale heterogeneous unlabeled cellular electron microscopy image dataset for deep learning. 96%
- Deep Neural Networks to Register and Annotate Cells in Moving and Deforming Nervous Systems 96%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.