ML Anthology

Gu, Zihan

1 publications

ICLR 2026 Deconstructing Positional Information: From Attention Logits to Training Biases Zihan Gu, Ruoyu Chen, Han Zhang, Hua Zhang, Yue Hu