Never Train from Scratch: Fair Comparison of Long-Sequence Models Requires Data-Driven Priors.

Ido Amos, Jonathan Berant, Ankit Gupta

Browse the full ICLR paper archive.