Image de couverture de livre

Specialization of softmax attention heads: insights from the high-dimensional single-location model

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Sagitova, M., Duranthon, O., Zdeborová, L.
Format:	Preprint
Publié:	2026
Sujets:	Machine Learning Disordered Systems and Neural Networks
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

Asymptotic generalization error of a single-layer graph convolutional network
par: Duranthon, O., et autres
Publié: (2024)

Statistical physics analysis of graph neural networks: Approaching optimality in the contextual stochastic block model
par: Duranthon, O., et autres
Publié: (2025)

Statistical Advantage of Softmax Attention: Insights from Single-Location Regression
par: Duranthon, O., et autres
Publié: (2025)

Fundamental limits of learning in sequence multi-index models and deep attention networks: High-dimensional asymptotics and sharp thresholds
par: Troiani, Emanuele, et autres
Publié: (2025)

Optimal thresholds and algorithms for a model of multi-modal learning in high dimensions
par: Keup, Christian, et autres
Publié: (2024)

High-dimensional Asymptotics of Denoising Autoencoders
par: Cui, Hugo, et autres
Publié: (2023)

Bayes optimal learning of attention-indexed models
par: Boncoraglio, Fabrizio, et autres
Publié: (2025)

On the existence of consistent adversarial attacks in high-dimensional linear classification
par: Vilucchio, Matteo, et autres
Publié: (2025)

A solvable high-dimensional model where nonlinear autoencoders learn structure invisible to PCA while test loss misaligns with generalization
par: Mendes, Vicente Conde, et autres
Publié: (2026)

Bilinear Sequence Regression: A Model for Learning from Long Sequences of High-dimensional Tokens
par: Erba, Vittorio, et autres
Publié: (2024)

Fundamental computational limits of weak learnability in high-dimensional multi-index models
par: Troiani, Emanuele, et autres
Publié: (2024)

Building Conformal Prediction Intervals with Approximate Message Passing
par: Clarté, Lucas, et autres
Publié: (2024)

Analysis of Bootstrap and Subsampling in High-dimensional Regularized Regression
par: Clarté, Lucas, et autres
Publié: (2024)

The Rules-and-Facts Model for Simultaneous Generalization and Memorization in Neural Networks
par: Farné, Gabriele, et autres
Publié: (2026)

Learning with Restricted Boltzmann Machines: Asymptotics of AMP and GD in High Dimensions
par: Xu, Yizhou, et autres
Publié: (2025)

Computational Thresholds in Multi-Modal Learning via the Spiked Matrix-Tensor Model
par: Tabanelli, Hugo, et autres
Publié: (2025)

Generative modeling through internal high-dimensional chaotic activity
par: Fournier, Samantha J., et autres
Publié: (2024)

Asymptotics of SGD in Sequence-Single Index Models and Single-Layer Attention Networks
par: Arnaboldi, Luca, et autres
Publié: (2025)

A solvable model of learning generative diffusion: theory and insights
par: Cui, Hugo, et autres
Publié: (2025)

Spectral Thresholds in Correlated Spiked Models and Fundamental Limits of Partial Least Squares
par: Mergny, Pierre, et autres
Publié: (2025)

Scaling and renormalization in high-dimensional regression
par: Atanasov, Alexander, et autres
Publié: (2024)

Asymptotic theory of in-context learning by linear attention
par: Lu, Yue M., et autres
Publié: (2024)

Asymptotics of feature learning in two-layer networks after one gradient-step
par: Cui, Hugo, et autres
Publié: (2024)

Rigorous Asymptotics for First-Order Algorithms Through the Dynamical Cavity Method
par: Dandi, Yatin, et autres
Publié: (2026)

Infinite Limits of Multi-head Transformer Dynamics
par: Bordelon, Blake, et autres
Publié: (2024)

In-context denoising with one-layer transformers: connections between attention and associative memory retrieval
par: Smart, Matthew, et autres
Publié: (2025)

Self-attention as an attractor network: transient memories without backpropagation
par: D'Amico, Francesco, et autres
Publié: (2024)

Sampling with flows, diffusion and autoregressive neural networks: A spin-glass perspective
par: Ghio, Davide, et autres
Publié: (2023)

The Nuclear Route: Sharp Asymptotics of ERM in Overparameterized Quadratic Networks
par: Erba, Vittorio, et autres
Publié: (2025)

Sharp description of local minima in the loss landscape of high-dimensional two-layer ReLU neural networks
par: Huang, Jie, et autres
Publié: (2026)

High-dimensional manifold of solutions in neural networks: insights from statistical physics
par: Malatesta, Enrico M.
Publié: (2023)

Quenches in the Sherrington-Kirkpatrick model
par: Erba, Vittorio, et autres
Publié: (2024)

Single-Head Attention in High Dimensions: A Theory of Generalization, Weights Spectra, and Scaling Laws
par: Boncoraglio, Fabrizio, et autres
Publié: (2025)

High-dimensional learning of narrow neural networks
par: Cui, Hugo
Publié: (2024)

Fundamental Limits of Matrix Sensing: Exact Asymptotics, Universality, and Applications
par: Xu, Yizhou, et autres
Publié: (2025)

Bayes-optimal learning of an extensive-width neural network from quadratically many samples
par: Maillard, Antoine, et autres
Publié: (2024)

Inference in Spreading Processes with Neural-Network Priors
par: Ghio, Davide, et autres
Publié: (2025)

Gaussian Universality of Perceptrons with Random Labels
par: Gerace, Federica, et autres
Publié: (2022)

Dynamical Cavity Method for Hypergraphs and its Application to Quenches in the k-XOR-SAT Problem
par: Maier, Aude, et autres
Publié: (2024)

Dynamical Phase Transitions in Graph Cellular Automata
par: Behrens, Freya, et autres
Publié: (2023)