English
Related papers

Related papers: Attention Sinks Are Provably Necessary in Softmax …

200 papers

Large Language Models (LLMs) often assign disproportionate attention to the first token, a phenomenon known as the attention sink. Several recent approaches aim to address this issue, including Sink Attention in GPT-OSS and Gated Attention…

Computation and Language · Computer Science 2026-05-28 Zizhuo Fu , Wenxuan Zeng , Runsheng Wang , Meng Li

The goal of this paper is to strengthen the reasoning of Omnimodal Large Language Models (Omni-LLMs) at inference time, without additional training. These models jointly process video, audio, and text, and given the large number of tokens…

Computer Vision and Pattern Recognition · Computer Science 2026-05-08 Suho Yoo , Youngjoon Jang , Joon Son Chung

Recently, it has been argued that encoder-decoder models can be made more interpretable by replacing the softmax function in the attention with its sparse variants. In this work, we introduce a novel, simple method for achieving sparsity in…

Computation and Language · Computer Science 2021-10-07 Biao Zhang , Ivan Titov , Rico Sennrich

The self-attention mechanism is central to the success of Transformer architectures. However, standard row-stochastic attention has been shown to suffer from significant signal degradation across layers. In particular, it can induce rank…

Machine Learning · Computer Science 2026-04-10 Michela Lapenna , Rita Fioresi , Bahman Gharesifard

Various forms of sparse attention have been explored to mitigate the quadratic computational and memory cost of the attention mechanism in transformers. We study sparse transformers not through a lens of efficiency but rather in terms of…

Machine Learning · Computer Science 2025-06-19 Parikshit Ram , Kenneth L. Clarkson , Tim Klinger , Shashanka Ubaru , Alexander G. Gray

Large transformer models have achieved state-of-the-art results in numerous natural language processing tasks. Among the pivotal components of the transformer architecture, the attention mechanism plays a crucial role in capturing token…

Computation and Language · Computer Science 2026-03-16 Yichuan Deng , Zhao Song , Kaijun Yuan , Tianyi Zhou

Diffusion Language Models (DLMs) have emerged as a compelling alternative to autoregressive approaches, enabling parallel text generation with competitive performance. Despite these advantages, there is a critical instability in DLMs: the…

Computation and Language · Computer Science 2026-02-24 Zihou Zhang , Zheyong Xie , Li Zhong , Haifeng Liu , Yao Hu , Shaosheng Cao

While transformer models exhibit strong in-context learning (ICL) abilities, they often fail to generalize under simple distribution shifts. We analyze these failures and identify Softmax, the scoring function in the attention mechanism, as…

Computation and Language · Computer Science 2026-05-12 Omar Naim , Swarnadeep Bhar , Jérôme Bolte , Nicholas Asher

Although transformer-based models have shown exceptional empirical performance, the fundamental principles governing their training dynamics are inadequately characterized beyond configuration-specific studies. Inspired by empirical…

Machine Learning · Computer Science 2025-10-09 Zheng-An Chen , Tao Luo

The dot product self-attention is known to be central and indispensable to state-of-the-art Transformer models. But is it really required? This paper investigates the true importance and contribution of the dot product-based self-attention…

Computation and Language · Computer Science 2021-05-25 Yi Tay , Dara Bahri , Donald Metzler , Da-Cheng Juan , Zhe Zhao , Che Zheng

A striking property of transformers is their ability to perform in-context learning (ICL), a machine learning framework in which the learner is presented with a novel context during inference implicitly through some data, and tasked with…

Machine Learning · Computer Science 2024-05-29 Liam Collins , Advait Parulekar , Aryan Mokhtari , Sujay Sanghavi , Sanjay Shakkottai

Multi-head attention enables transformer models to represent multiple attention patterns simultaneously. Empirically, head specialization emerges in distinct stages during training, while many heads remain redundant and learn similar…

Machine Learning · Computer Science 2026-03-05 M. Sagitova , O. Duranthon , L. Zdeborová

While linear attention reduces the quadratic complexity of standard Transformers to linear time, it often lags behind in expressivity due to the removal of softmax normalization. This omission eliminates \emph{global competition}, a…

Machine Learning · Computer Science 2026-02-03 Mingwei Xu , Xuan Lin , Xinnan Guo , Wanqing Xu , Wanyun Cui

Self-attention is the key mechanism of transformers, which are the essential building blocks of modern foundation models. Recent studies have shown that pure self-attention suffers from an increasing degree of rank collapse as depth…

Machine Learning · Computer Science 2024-11-04 Xinyi Wu , Amir Ajorlou , Yifei Wang , Stefanie Jegelka , Ali Jadbabaie

Transformers are increasingly adopted for modeling and forecasting time-series, yet their internal mechanisms remain poorly understood from a dynamical systems perspective. In contrast to classical autoregressive and state-space models,…

Machine Learning · Computer Science 2025-12-25 Gregory Duthé , Nikolaos Evangelou , Wei Liu , Ioannis G. Kevrekidis , Eleni Chatzi

Large Language Models (LLMs) tend to attend heavily to the first token in the sequence -- creating a so-called attention sink. Many works have studied this phenomenon in detail, proposing various ways to either leverage or alleviate it.…

Transformers are extremely successful machine learning models whose mathematical properties remain poorly understood. Here, we rigorously characterize the behavior of transformers with hardmax self-attention and normalization sublayers as…

Computation and Language · Computer Science 2026-05-14 Albert Alcalde , Giovanni Fantuzzi , Enrique Zuazua

Softmax-based dot-product attention is a cornerstone of Transformer architectures, enabling remarkable capabilities such as in-context learning. However, as context lengths increase, a fundamental limitation of the softmax function emerges:…

Machine Learning · Computer Science 2026-02-12 Sai Surya Duvvuri , Nirmal Patel , Nilesh Gupta , Inderjit S. Dhillon

The Transformer model architecture has become one of the most widely used in deep learning and the attention mechanism is at its core. The standard attention formulation uses a softmax operation applied to a scaled dot product between query…

Machine Learning · Computer Science 2026-04-02 Hariprasath Govindarajan , Per Sidén , Jacob Roll , Fredrik Lindsten

Attention sink (AS) is a consistent pattern in transformer attention maps where certain tokens (often special tokens or positional anchors) disproportionately attract attention from other tokens. We show that in transformers, AS is not an…

Machine Learning · Computer Science 2025-08-05 Valeria Ruscio , Umberto Nanni , Fabrizio Silvestri