Jul 2026
Interior interpretability with attention rollout: contraction and propagation profiles in Transformers
Attention rollout is used here as a diagnostic of attention-mediated propagation, not as a causal explanation or faithful attribution of the complete Transformer, but as a structural interpretation to the corresponding rollout propagation profile.
Umberto Biccari, Qianqi Huang, Enrique Zuazua
· arXiv.org · 0 citations