Cover Image

Two failure modes of deep transformers and how to avoid them: a unified theory of signal propagation at initialisation

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Giorlandino, Alessio, Goldt, Sebastian
Format:	Preprint
Published:	2025
Subjects:	Machine Learning Disordered Systems and Neural Networks Statistical Mechanics
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Factual recall in linear associative memories: sharp asymptotics and mechanistic insights
by: Giorlandino, Alessio, et al.
Published: (2026)

Memorisation, convergence and generalisation in generative models
by: Maillard, Antoine, et al.
Published: (2026)

A Fourier perspective on the learning dynamics of neural networks: from sample complexities to mechanistic insights
by: Ricci, Fabiola, et al.
Published: (2026)

Feature learning from non-Gaussian inputs: the case of Independent Component Analysis in high dimensions
by: Ricci, Fabiola, et al.
Published: (2025)

Mapping of attention mechanisms to a generalized Potts model
by: Rende, Riccardo, et al.
Published: (2023)

Spring-block theory of feature learning in deep neural networks
by: Shi, Cheng, et al.
Published: (2024)

Distinct mechanisms underlying in-context learning in transformers
by: Gibson, Cole, et al.
Published: (2026)

Generalization Dynamics of Linear Diffusion Models
by: Merger, Claudia, et al.
Published: (2025)

Analytic theory of dropout regularization
by: Mori, Francesco, et al.
Published: (2025)

Dynamic neuron approach to deep neural networks: Decoupling neurons for renormalization group analysis
by: Lee, Donghee, et al.
Published: (2024)

Weight fluctuations in (deep) linear neural networks and a derivation of the inverse-variance flatness relation
by: Gross, Markus, et al.
Published: (2023)

Fundamental operating regimes, hyper-parameter fine-tuning and glassiness: towards an interpretable replica-theory for trained restricted Boltzmann machines
by: Fachechi, Alberto, et al.
Published: (2024)

Self-consistent theory for sound propagation in a simple model of a disordered, harmonic solid
by: Szamel, Grzegorz
Published: (2024)

A theoretical perspective on mode collapse in variational inference
by: Soletskyi, Roman, et al.
Published: (2024)

Nonlinear dielectric response in glasses: restoring forces and avoided spin-glass criticality
by: Bertin, Eric, et al.
Published: (2023)

Application of deep neural networks for computing the renormalization group flow of the two-dimensional phi^4 field theory
by: Zhao, Yueqi, et al.
Published: (2025)

Compression theory for inhomogeneous systems
by: Gökmen, Doruk Efe, et al.
Published: (2023)

The Copycat Perceptron: Smashing Barriers Through Collective Learning
by: Catania, Giovanni, et al.
Published: (2023)

The autoregressive neural network architecture of the Boltzmann distribution of pairwise interacting spins systems
by: Biazzo, Indaco
Published: (2023)

Interpreting the Synchronization Gap: The Hidden Mechanism Inside Diffusion Transformers
by: Albrychiewicz, Emil, et al.
Published: (2026)

Optimal Protocols for Continual Learning via Statistical Physics and Control Theory
by: Mori, Francesco, et al.
Published: (2024)

Universal Scaling Laws of Absorbing Phase Transitions in Artificial Deep Neural Networks
by: Tamai, Keiichi, et al.
Published: (2023)

Inferring Higher-Order Couplings with Neural Networks
by: Decelle, Aurélien, et al.
Published: (2025)

Explaining the effects of non-convergent sampling in the training of Energy-Based Models
by: Agoritsas, Elisabeth, et al.
Published: (2023)

Discrete generative diffusion models without stochastic differential equations: a tensor network approach
by: Causer, Luke, et al.
Published: (2024)

The Physics of Data and Tasks: Theories of Locality and Compositionality in Deep Learning
by: Favero, Alessandro
Published: (2025)

Sampling with flows, diffusion and autoregressive neural networks: A spin-glass perspective
by: Ghio, Davide, et al.
Published: (2023)

Stochastic Gradient Descent-like relaxation is equivalent to Metropolis dynamics in discrete optimization and inference problems
by: Angelini, Maria Chiara, et al.
Published: (2023)

Escape dynamics and implicit bias of one-pass SGD in overparameterized quadratic networks
by: Bocchi, Dario, et al.
Published: (2026)

Inferring effective couplings with Restricted Boltzmann Machines
by: Decelle, Aurélien, et al.
Published: (2023)

From Classical to Quantum: Extending Prometheus for Unsupervised Discovery of Phase Transitions in Three Dimensions and Quantum Systems
by: Yee, Brandon, et al.
Published: (2026)

Transfer Learning in $\ell_1$ Regularized Regression: Hyperparameter Selection Strategy based on Sharp Asymptotic Analysis
by: Okajima, Koki, et al.
Published: (2024)

A theoretical framework for overfitting in energy-based modeling
by: Catania, Giovanni, et al.
Published: (2025)

Uncertainty in AI-driven Monte Carlo simulations
by: Tzivrailis, Dimitrios, et al.
Published: (2025)

Generalization vs. Specialization under Concept Shift
by: Nguyen, Alex, et al.
Published: (2024)

Gaussian Universality in Neural Network Dynamics with Generalized Structured Input Distributions
by: Bae, Jaeyong, et al.
Published: (2024)

Coding schemes in neural networks learning classification tasks
by: van Meegen, Alexander, et al.
Published: (2024)

Nonequilbrium physics of generative diffusion models
by: Yu, Zhendong, et al.
Published: (2024)

The Symmetric Perceptron: a Teacher-Student Scenario
by: Catania, Giovanni, et al.
Published: (2026)

Pseudo-likelihood produces associative memories able to generalize, even for asymmetric couplings
by: D'Amico, Francesco, et al.
Published: (2025)