How saccadic vision might help with theinterpretability of deep networks

Iana Sereda, Grigory Osipov

We describe how some problems (interpretability,lack of object-orientedness) of modern deep networks potentiallycould be solved by adapting a biologically plausible saccadicmechani…