Infinite attention: NNGP and NTK for deep attention networks.

Jiri Hron, Yasaman Bahri, Jascha Sohl-Dickstein, Roman Novak

Browse the full ICML paper archive.