From Shortcut to Induction Head: How Data Diversity Shapes Algorithm Selection in Transformers

NeurIPSSpotlight2025

Authors: Ryotaro Kawata, Yujin Song, Alberto Bietti, Naoki Nishikawa, Taiji Suzuki, Samuel Vaiter, Denny Wu
Venue: NeurIPS 2025
Track: Spotlight

TL;DR

Transformers can implement both generalizable algorithms (e…

Opening excerpt from the authors’ abstract. source

Read the paper

OpenReview

Topics

transformer

← All NeurIPS 2025 Spotlight papers · Browse the whole archive