English
Related papers

Related papers: Breaking Symmetry When Training Transformers

200 papers

Transformer models are permutation equivariant. To supply the order and type information of the input tokens, position and segment embeddings are usually added to the input. Recent works proposed variations of positional encodings with…

Computation and Language · Computer Science 2021-11-04 Pu-Chin Chen , Henry Tsai , Srinadh Bhojanapalli , Hyung Won Chung , Yin-Wen Chang , Chun-Sung Ferng

Since its introduction, the transformer has shifted the development trajectory away from traditional models (e.g., RNN, MLP) in time series forecasting, which is attributed to its ability to capture global dependencies within temporal…

Machine Learning · Computer Science 2025-01-07 Xiwen Chen , Peijie Qiu , Wenhui Zhu , Huayu Li , Hao Wang , Aristeidis Sotiras , Yalin Wang , Abolfazl Razi

The Transformer architecture has become prominent in developing large causal language models. However, mechanisms to explain its capabilities are not well understood. Focused on the training process, here we establish a meta-learning view…

Machine Learning · Computer Science 2024-03-26 Xinbo Wu , Lav R. Varshney

As transformers are equivariant to the permutation of input tokens, encoding the positional information of tokens is necessary for many tasks. However, since existing positional encoding schemes have been initially designed for NLP tasks,…

Computer Vision and Pattern Recognition · Computer Science 2024-06-10 Takeru Miyato , Bernhard Jaeger , Max Welling , Andreas Geiger

Recent advancements in transformer-based models have greatly improved time series analysis, providing robust solutions for tasks such as forecasting, anomaly detection, and classification. A crucial element of these models is positional…

Machine Learning · Computer Science 2026-05-07 Habib Irani , Vangelis Metsis

Systematic compositionality is an essential mechanism in human language, allowing the recombination of known parts to create novel expressions. However, existing neural models have been shown to lack this basic ability in learning symbolic…

Computation and Language · Computer Science 2021-10-01 Yichen Jiang , Mohit Bansal

Understanding the transformer architecture and its workings is essential for machine learning (ML) engineers. However, truly understanding the transformer architecture can be demanding, even if you have a solid background in machine…

Machine Learning · Computer Science 2025-02-28 Joni-Kristian Kämäräinen

In order to preserve word-order information in a non-autoregressive setting, transformer architectures tend to include positional knowledge, by (for instance) adding positional encodings to token embeddings. Several modifications have been…

Computation and Language · Computer Science 2021-09-14 Vinit Ravishankar , Anders Søgaard

This paper investigates the dynamical properties of tokens in pre-trained Transformer models and explores their application to improving Transformers. To this end, we analyze the dynamical system governing the continuous-time limit of the…

Machine Learning · Computer Science 2025-12-04 Duy-Tung Pham , An The Nguyen , Viet-Hoang Tran , Nhan-Phu Chung , Xin T. Tong , Tan M. Nguyen , Thieu N. Vo

When trained on tasks requiring an understanding of hierarchical structure, transformers have been found to represent this hierarchy in distinct ways: in the geometry of the residual stream, and in stack-like attention patterns maintaining…

Computation and Language · Computer Science 2026-04-27 Aryan Sharma , Cutter Dawes , Shivam Raval

Attentional mechanisms are order-invariant. Positional encoding is a crucial component to allow attention-based deep model architectures such as Transformer to address sequences or images where the position of information matters. In this…

Machine Learning · Computer Science 2021-11-10 Yang Li , Si Si , Gang Li , Cho-Jui Hsieh , Samy Bengio

With the attention mechanism, transformers achieve significant empirical successes. Despite the intuitive understanding that transformers perform relational inference over long sequences to produce desirable representations, we lack a…

Machine Learning · Computer Science 2024-04-02 Yufeng Zhang , Boyi Liu , Qi Cai , Lingxiao Wang , Zhaoran Wang

Transformers are being used extensively across several sequence modeling tasks. Significant research effort has been devoted to experimentally probe the inner workings of Transformers. However, our conceptual and theoretical understanding…

Machine Learning · Computer Science 2020-10-13 Satwik Bhattamishra , Arkil Patel , Navin Goyal

We reveal that transformers trained in an autoregressive manner naturally encode time-delayed causal structures in their learned representations. When predicting future values in multivariate time series, the gradient sensitivities of…

Machine Learning · Computer Science 2026-01-12 Xinyue Wang , Stephen Wang , Biwei Huang

While Transformers have rapidly gained popularity in various computer vision applications, post-hoc explanations of their internal mechanisms remain largely unexplored. Vision Transformers extract visual information by representing image…

Computer Vision and Pattern Recognition · Computer Science 2024-03-22 Junyi Wu , Bin Duan , Weitai Kang , Hao Tang , Yan Yan

Positional encodings enable Transformers to incorporate sequential information, yet their theoretical understanding remains limited to two properties: distance attenuation and translation invariance. Because natural language lacks purely…

Machine Learning · Computer Science 2026-02-11 Zihan Gu , Ruoyu Chen , Han Zhang , Hua Zhang , Yue Hu

There is a growing interest in the ability of neural networks to execute algorithmic tasks (e.g., arithmetic, summary statistics, and sorting). The goal of this work is to better understand the role of attention in Transformers for…

Machine Learning · Computer Science 2025-06-11 Artur Back de Luca , George Giapitzakis , Shenghao Yang , Petar Veličković , Kimon Fountoulakis

Positional encoding is essential for supplementing transformer with positional information of tokens. Existing positional encoding methods demand predefined token/feature order, rendering them unsuitable for real-world data with…

Machine Learning · Computer Science 2025-09-25 Kaichen Xu , Yihang Du , Mianpeng Liu , Zimu Yu , Xiaobo Sun

Transformer-based models have brought a radical change to neural machine translation. A key feature of the Transformer architecture is the so-called multi-head attention mechanism, which allows the model to focus simultaneously on different…

Computation and Language · Computer Science 2020-10-06 Alessandro Raganato , Yves Scherrer , Jörg Tiedemann

Transformer architectures rely on explicit position encodings in order to preserve a notion of word order. In this paper, we argue that existing work does not fully utilize position information. For example, the initial proposal of a…

Computation and Language · Computer Science 2020-09-30 Zhiheng Huang , Davis Liang , Peng Xu , Bing Xiang