Models of primate ventral stream that categorize and visualize images

We show that computational models that identify objects while also retaining enough information about the input image to allow its reconstruction, can provide a better description of how primate VS represents this category-orthogonal information, than do models that “just” identify objects

[PDF]

Published: