arXiv · 2611.18088
Two-Clustering Regime of Token Dynamics in Causal Attention
Abstract
We partially resolve Conjecture 1 of [Karagodin et al., Clustering in Causal Attention Masking (NeurIPS 2024)] and extend it to time-dependent parameter regimes. The conjecture predicts that causal attention dynamics converge to two clusters aligned with the eigenvectors associated with the largest eigenvalue $λ_{\max}$ of the value transformation. We prove that this behavior holds when $λ_{\max}>0$ and is generally false when $λ_{\max}\le0$. The resulting two-cluster regime is common in practice and substantially more challenging than the classical single-cluster setting studied previously. A central contribution of our work is the identification of a rigorous dimension reduction phenomenon: despite evolving in a high-dimensional embedding space, token trajectories collapse onto a two-dimensional structure governed by a small number of dominant directions. Building on this insight, we introduce a new analytical framework--instability within the flow--inspired by instability phenomena in physical systems. This framework enables the analysis of steady-regime behavior in non-monotone, non-autonomous attention dynamics. Our theoretical findings are illustrated through simulations and experimental studies using Group Query Attention (GQA) layers.
Explore related subjects
Keep this discovery
Explore connections, maps & timelines
Trinh T. Nguyen, Duy-Tung Pham, Hoang-Son Do, Tan Minh Nguyen, Thieu N. Vo. 2026-11-13. Two-Clustering Regime of Token Dynamics in Causal Attention. https://arxiv.science/abs/2611.18088
Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite arXiv Science by Indiciae: BibTeX · RIS.