A mean-field kinetic theory derivation produces a closed-form U-shaped token retrieval profile that explains the lost-in-the-middle phenomenon in Transformers.
Quantitative Clustering in Mean-Field Transformer Models
7 Pith papers cite this work. Polarity classification is still indexing.
7
Pith papers citing it
abstract
The evolution of tokens through deep transformer models can be modeled as an interacting particle system that has been shown to exhibit an asymptotic clustering behavior akin to the synchronization phenomenon in Kuramoto models. In this work, we investigate the long-time clustering of mean-field transformer models. More precisely, under suitable assumptions on the transformer model parameters, we establish that any suitably regular mean-field initialization synchronizes exponentially fast to a Dirac point mass, with explicit quantitative convergence rates.
citation-role summary
background 2
citation-polarity summary
years
2026 7roles
background 2polarities
background 2representative citing papers
Transformers converge pathwise to a stochastic particle system and SPDE in the scaling limit, exhibiting synchronization by noise and exponential energy dissipation when common noise is coercive relative to self-attention drift.
Symmetric self-attention dynamics select the dominant eigendirection of V, producing homogeneous alignment when one positive eigenvalue dominates or sign-split polarization when V is negative definite.
Derives forward and backward propagation-of-chaos bounds for finite vs. infinite-context transformers modeled as contextual flow maps, achieving Wasserstein rate n^{-1/d} generally and n^{-1/2} for transformer-like cases.
Models multi-head transformer data flow as time-dependent Wasserstein gradient flows of an attention-capturing interaction energy, with proofs on omega-limit stationary points and stability under weight and input perturbations.
In the low-temperature regime, the token distribution in mean-field transformers concentrates onto the push-forward under a key-query-value projection with Wasserstein distance scaling as √(log(β+1)/β) exp(Ct) + exp(-ct).
Multi-head self-attention dynamics admit a non-decreasing energy functional under suitable score-matrix conditions, with closed-form clustering thresholds and monotonic entropy production in simplified regimes.
citing papers explorer
-
Kinetic theory for Transformers and the lost-in-the-middle phenomenon
A mean-field kinetic theory derivation produces a closed-form U-shaped token retrieval profile that explains the lost-in-the-middle phenomenon in Transformers.
-
Stochastic Scaling Limits and Synchronization by Noise in Deep Transformer Models
Transformers converge pathwise to a stochastic particle system and SPDE in the scaling limit, exhibiting synchronization by noise and exponential energy dissipation when common noise is coercive relative to self-attention drift.
-
Spectral Selection in Symmetric Self-Attention Dynamics
Symmetric self-attention dynamics select the dominant eigendirection of V, producing homogeneous alignment when one positive eigenvalue dominates or sign-split polarization when V is negative definite.
-
Propagation of Chaos in Contextual Flow Maps
Derives forward and backward propagation-of-chaos bounds for finite vs. infinite-context transformers modeled as contextual flow maps, achieving Wasserstein rate n^{-1/d} generally and n^{-1/2} for transformer-like cases.
-
Multi-Headed Transformer Architectures as Time-dependent Wasserstein Gradient Flows
Models multi-head transformer data flow as time-dependent Wasserstein gradient flows of an attention-capturing interaction energy, with proofs on omega-limit stationary points and stability under weight and input perturbations.
-
Quantifying Concentration Phenomena of Mean-Field Transformers in the Low-Temperature Regime
In the low-temperature regime, the token distribution in mean-field transformers concentrates onto the push-forward under a key-query-value projection with Wasserstein distance scaling as √(log(β+1)/β) exp(Ct) + exp(-ct).
-
Gradient Flow Structure and Quantitative Dynamics of Multi-Head Self-Attention
Multi-head self-attention dynamics admit a non-decreasing energy functional under suitable score-matrix conditions, with closed-form clustering thresholds and monotonic entropy production in simplified regimes.