Kinetic theory for Transformers and the lost-in-the-middle phenomenon

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Duerinckx, Mitia, Geshkovski, Borjan, Rossi, Stefano
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914548589527040
author Duerinckx, Mitia
Geshkovski, Borjan
Rossi, Stefano
author_facet Duerinckx, Mitia
Geshkovski, Borjan
Rossi, Stefano
contents We study causal self-attention dynamics -- a toy model for decoder Transformers -- which we interpret as a non-exchangeable interacting particle system. Adapting cumulant expansions to the triangular causal dependency structure of the model, and appealing to non-hierarchical methods to estimate correlations using Glauber calculus, we prove a quantitative mean-field limit result and a next-order characterization of correlations. For iid uniformly distributed tokens, the limiting correlation equation can be solved in closed form and we obtain a rigorous explanation of the empirically observed \emph{lost-in-the-middle} phenomenon: the token retrieval profile, as a function of the source position in the prompt, is $\mathsf{U}$-shaped, with primacy, recency, and a unique interior minimum under an explicit smallness condition.
format Preprint
id arxiv_https___arxiv_org_abs_2605_09213
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Kinetic theory for Transformers and the lost-in-the-middle phenomenon
Duerinckx, Mitia
Geshkovski, Borjan
Rossi, Stefano
Analysis of PDEs
Machine Learning
Probability
We study causal self-attention dynamics -- a toy model for decoder Transformers -- which we interpret as a non-exchangeable interacting particle system. Adapting cumulant expansions to the triangular causal dependency structure of the model, and appealing to non-hierarchical methods to estimate correlations using Glauber calculus, we prove a quantitative mean-field limit result and a next-order characterization of correlations. For iid uniformly distributed tokens, the limiting correlation equation can be solved in closed form and we obtain a rigorous explanation of the empirically observed \emph{lost-in-the-middle} phenomenon: the token retrieval profile, as a function of the source position in the prompt, is $\mathsf{U}$-shaped, with primacy, recency, and a unique interior minimum under an explicit smallness condition.
title Kinetic theory for Transformers and the lost-in-the-middle phenomenon
topic Analysis of PDEs
Machine Learning
Probability
url https://arxiv.org/abs/2605.09213