The Asymptotic Behavior of Attention in Transformers

• URL: http://arxiv.org/abs/2412.02682v2
• Date: Thu, 25 Sep 2025 00:09:18 GMT
• Title: The Asymptotic Behavior of Attention in Transformers
• Authors: Álvaro Rodríguez Abella, João Pedro Silvestre, Paulo Tabuada,
• Abstract summary: The transformer architecture has become the foundation of modern Large Language Models (LLMs)<n>We show that all the tokens in a transformerally converge to a cluster as depth increases.<n>We then extend our analysis to autoregressive models, exploiting their structure to further generalize the theoretical guarantees.
• Score: 2.631744051718347
• License: http://creativecommons.org/licenses/by-nc-nd/4.0/
• Abstract: The transformer architecture has become the foundation of modern Large Language Models (LLMs), yet its theoretical properties are still not well understood. As with classic neural networks, a common approach to improve these models is to increase their size and depth. However, such strategies may be suboptimal, as several works have shown that adding more layers yields increasingly diminishing returns. More importantly, prior studies have shown that increasing depth may lead to model collapse, i.e., all the tokens converge to a single cluster, undermining the ability of LLMs to generate diverse outputs. Building on differential equation models for the transformer dynamics, we prove that all the tokens in a transformer asymptotically converge to a cluster as depth increases. At the technical level we leverage tools from control theory, including consensus dynamics on manifolds and input-to-state stability (ISS). We then extend our analysis to autoregressive models, exploiting their structure to further generalize the theoretical guarantees.

Related papers

• Rethinking the Role of Dynamic Sparse Training for Scalable Deep Reinforcement Learning [58.533203990515034]
  Scaling neural networks has driven breakthrough advances in machine learning, yet this paradigm fails in deep reinforcement learning (DRL)<n>We show that dynamic sparse training strategies provide module-specific benefits that complement the primary scalability foundation established by architectural improvements.<n>We finally distill these insights into Module-Specific Training (MST), a practical framework that exploits the benefits of architectural improvements and demonstrates substantial scalability gains across diverse RL algorithms without algorithmic modifications.
  arXiv  Detail & Related papers  (2025-10-14T03:03:08Z)
• Universal Approximation Theorem for a Single-Layer Transformer [0.0]
  Deep learning employs multi-layer neural networks trained via the backpropagation algorithm.<n>Transformers have achieved state-of-the-art performance in natural language processing.<n>We prove that a single-layer Transformer, comprising one self-attention layer followed by a position-wise feed-forward network with ReLU activation, can any continuous sequence-to-sequence mapping on a compact domain to arbitrary precision.
  arXiv  Detail & Related papers  (2025-07-11T11:37:39Z)
• Growing Transformers: Modular Composition and Layer-wise Expansion on a Frozen Substrate [1.0152838128195467]
  The prevailing paradigm for scaling large language models (LLMs) involves monolithic, end-to-end training.<n>This paper explores an alternative, constructive scaling paradigm, enabled by the principle of emergent semantics in Transformers.<n>We operationalize this with a layer-wise constructive methodology that combines strict layer freezing in early stages with efficient, holistic fine-tuning of the entire model stack.
  arXiv  Detail & Related papers  (2025-07-08T20:01:15Z)
• Transformers as Multi-task Learners: Decoupling Features in Hidden Markov Models [12.112842686827669]
  Transformer based models have shown remarkable capabilities in sequence learning across a wide range of tasks.<n>We investigate the layerwise behavior of Transformers to uncover the mechanisms underlying their multi-task generalization ability.<n>Our explicit constructions align closely with empirical observations, providing theoretical support for the Transformer's effectiveness and efficiency on sequence learning across diverse tasks.
  arXiv  Detail & Related papers  (2025-06-02T17:39:31Z)
• Model Hemorrhage and the Robustness Limits of Large Language Models [119.46442117681147]
  Large language models (LLMs) demonstrate strong performance across natural language processing tasks, yet undergo significant performance degradation when modified for deployment.<n>We define this phenomenon as model hemorrhage - performance decline caused by parameter alterations and architectural changes.
  arXiv  Detail & Related papers  (2025-03-31T10:16:03Z)
• Neural ODE Transformers: Analyzing Internal Dynamics and Adaptive Fine-tuning [30.781578037476347]
  We introduce a novel approach to modeling transformer architectures using highly flexible non-autonomous neural ordinary differential equations (ODEs)<n>Our proposed model parameterizes all weights of attention and feed-forward blocks through neural networks, expressing these weights as functions of a continuous layer index.<n>Our neural ODE transformer demonstrates performance comparable to or better than vanilla transformers across various configurations and datasets.
  arXiv  Detail & Related papers  (2025-03-03T09:12:14Z)
• A Unified Perspective on the Dynamics of Deep Transformers [24.094975798576783]
  We study the evolution of data anisotropy through a deep Transformer. We highlight a clustering phenomenon that parallels previous results in the non-normalized discrete case.
  arXiv  Detail & Related papers  (2025-01-30T13:04:54Z)
• Interpreting Affine Recurrence Learning in GPT-style Transformers [54.01174470722201]
  In-context learning allows GPT-style transformers to generalize during inference without modifying their weights. This paper focuses specifically on their ability to learn and predict affine recurrences as an ICL task. We analyze the model's internal operations using both empirical and theoretical approaches.
  arXiv  Detail & Related papers  (2024-10-22T21:30:01Z)
• A Theoretical Understanding of Chain-of-Thought: Coherent Reasoning and Error-Aware Demonstration [41.88275731297211]
  We show that, compared to Stepwise ICL, the transformer gains better error correction ability and more accurate predictions if the reasoning from earlier steps is integrated. We propose an improvement on CoT by incorporating both correct and incorrect reasoning paths in the demonstration.
  arXiv  Detail & Related papers  (2024-10-21T22:07:20Z)
• Clustering in pure-attention hardmax transformers and its role in sentiment analysis [0.0]
  We rigorously characterize the behavior of transformers with hardmax self-attention and normalization sublayers as the number of layers tends to infinity. We show that the transformer inputsally converge to a clustered equilibrium determined by special points called leaders. We then leverage this theoretical understanding to solve sentiment analysis problems from language processing using a fully interpretable transformer model.
  arXiv  Detail & Related papers  (2024-06-26T16:13:35Z)
• Dynamical Mean-Field Theory of Self-Attention Neural Networks [0.0]
  Transformer-based models have demonstrated exceptional performance across diverse domains. Little is known about how they operate or what are their expected dynamics. We use methods for the study of asymmetric Hopfield networks in nonequilibrium regimes.
  arXiv  Detail & Related papers  (2024-06-11T13:29:34Z)
• Understanding the Expressive Power and Mechanisms of Transformer for Sequence Modeling [10.246977481606427]
  We study the mechanisms through which different components of Transformer, such as the dot-product self-attention, affect its expressive power. Our study reveals the roles of critical parameters in the Transformer, such as the number of layers and the number of attention heads.
  arXiv  Detail & Related papers  (2024-02-01T11:43:13Z)
• Efficient Compression of Overparameterized Deep Models through Low-Dimensional Learning Dynamics [10.673414267895355]
  We present a novel approach for compressing over parameterized models. Our algorithm improves the training efficiency by more than 2x, without compromising generalization.
  arXiv  Detail & Related papers  (2023-11-08T23:57:03Z)
• On the Convergence of Encoder-only Shallow Transformers [62.639819460956176]
  We build the global convergence theory of encoder-only shallow Transformers under a realistic setting. Our results can pave the way for a better understanding of modern Transformers, particularly on training dynamics.
  arXiv  Detail & Related papers  (2023-11-02T20:03:05Z)
• Transformers are Universal Predictors [21.92580010179886]
  We find limits to the Transformer architecture for language modeling and show it has a universal prediction property in an information-theoretic sense. We analyze performance in non-asymptotic data regimes to understand the role of various components of the Transformer architecture, especially in the context of data-efficient training.
  arXiv  Detail & Related papers  (2023-07-15T16:19:37Z)
• Approximation and Estimation Ability of Transformers for Sequence-to-Sequence Functions with Infinite Dimensional Input [50.83356836818667]
  We study the approximation and estimation ability of Transformers as sequence-to-sequence functions with infinite dimensional inputs. Our theoretical results support the practical success of Transformers for high dimensional data.
  arXiv  Detail & Related papers  (2023-05-30T02:44:49Z)
• A Theoretical Understanding of Shallow Vision Transformers: Learning, Generalization, and Sample Complexity [71.11795737362459]
  ViTs with self-attention modules have recently achieved great empirical success in many tasks. However, theoretical learning generalization analysis is mostly noisy and elusive. This paper provides the first theoretical analysis of a shallow ViT for a classification task.
  arXiv  Detail & Related papers  (2023-02-12T22:12:35Z)
• DIFFormer: Scalable (Graph) Transformers Induced by Energy Constrained Diffusion [66.21290235237808]
  We introduce an energy constrained diffusion model which encodes a batch of instances from a dataset into evolutionary states. We provide rigorous theory that implies closed-form optimal estimates for the pairwise diffusion strength among arbitrary instance pairs. Experiments highlight the wide applicability of our model as a general-purpose encoder backbone with superior performance in various tasks.
  arXiv  Detail & Related papers  (2023-01-23T15:18:54Z)
• Outliers Dimensions that Disrupt Transformers Are Driven by Frequency [79.22656609637525]
  We show that the token frequency contributes to the outlier phenomenon. We also find that, surprisingly, the outlier effect on the model performance varies by layer, and that variance is also related to the correlation between outlier magnitude and encoded token frequency.
  arXiv  Detail & Related papers  (2022-05-23T15:19:09Z)
• A Probabilistic Interpretation of Transformers [91.3755431537592]
  We propose a probabilistic interpretation of exponential dot product attention of transformers and contrastive learning based off of exponential families. We state theoretical limitations of our theory and the Hopfield theory and suggest directions for resolution.
  arXiv  Detail & Related papers  (2022-04-28T23:05:02Z)
• On the Power of Saturated Transformers: A View from Circuit Complexity [87.20342701232869]
  We show that saturated transformers transcend the limitations of hard-attention transformers. The jump from hard to saturated attention can be understood as increasing the transformer's effective circuit depth by a factor of $O(log n)$.
  arXiv  Detail & Related papers  (2021-06-30T17:09:47Z)
• Reinforcement Learning as One Big Sequence Modeling Problem [84.84564880157149]
  Reinforcement learning (RL) is typically concerned with estimating single-step policies or single-step models. We view RL as a sequence modeling problem, with the goal being to predict a sequence of actions that leads to a sequence of high rewards.
  arXiv  Detail & Related papers  (2021-06-03T17:58:51Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.