Saved in:
| Main Authors: | , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2512.09275 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912979740524544 |
|---|---|
| author | He, Weiyi Xing, Yue |
| author_facet | He, Weiyi Xing, Yue |
| contents | Positional encoding (PE) is a core architectural component of Transformers, yet its impact on the Transformer's generalization and robustness remains unclear. In this work, we provide the first generalization analysis for a single-layer Transformer under in-context regression that explicitly accounts for a completely trainable PE module. Our result shows that PE systematically enlarges the generalization gap. Extending to the adversarial setting, we derive the adversarial Rademacher generalization bound. We find that the gap between models with and without PE is magnified under attack, demonstrating that PE amplifies the vulnerability of models. Our bounds are empirically validated by a simulation study. Together, this work establishes a new framework for understanding the clean and adversarial generalization in ICL with PE. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2512_09275 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Impact of Positional Encoding: Clean and Adversarial Rademacher Complexity for Transformers under In-Context Regression He, Weiyi Xing, Yue Machine Learning Positional encoding (PE) is a core architectural component of Transformers, yet its impact on the Transformer's generalization and robustness remains unclear. In this work, we provide the first generalization analysis for a single-layer Transformer under in-context regression that explicitly accounts for a completely trainable PE module. Our result shows that PE systematically enlarges the generalization gap. Extending to the adversarial setting, we derive the adversarial Rademacher generalization bound. We find that the gap between models with and without PE is magnified under attack, demonstrating that PE amplifies the vulnerability of models. Our bounds are empirically validated by a simulation study. Together, this work establishes a new framework for understanding the clean and adversarial generalization in ICL with PE. |
| title | Impact of Positional Encoding: Clean and Adversarial Rademacher Complexity for Transformers under In-Context Regression |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2512.09275 |