arXiv - CSCL: "Simplicity Bias in Transformers and their Ability…" - Qoto Mastodon

arXiv - CSCL @arxiv_cscl@qoto.org

Simplicity Bias in Transformers and their Ability to Learn Sparse Boolean Functions. (arXiv:2211.12316v2 [cs.LG] UPDATED)

http://arxiv.org/abs/2211.12316 #arXiv #NLProc

Jul 11, 2023, 03:24 · · arxiv-cscl · · ·

Sign in to participate in the conversation