Why Do Accumulated Transformations Extrapolate?
Mirrored from arXiv — NLP / Computation & Language for archival readability. Support the source by reading on the original site.
Computer Science > Machine Learning
Title:Why Do Accumulated Transformations Extrapolate?
Abstract:PaTH Attention showed that replacing RoPE's position-indexed rotations with accumulated data-dependent Householder reflections yields strong length extrapolation, though performance degrades at extreme context lengths. We ask whether this depends on Householder-specific structure or reflects a general property of accumulated transformations along source-to-query paths. We study a simpler variant keeping RoPE's block-diagonal SO(2) rotations but replacing position-indexed angles with accumulated token-dependent ones. It shows the same pattern: improved extrapolation then degradation at long contexts. We prove the result extends to accumulated orthogonal transformations satisfying certain regularity conditions: their products become incoherent after finitely many steps, suppressing attention to distant tokens. Accumulated rotations of queries and keys create a finite mixing window independent of context length; per-token suppression learned in training transfers unchanged to any evaluation length, and high-dimensional concentration produces a score gap suppressing far tokens while near-route transport preserves the target signal. Conversely, a lower bound shows accumulated rotations must eventually degrade: as the far set grows, no rotations preserve the near signal without explicit far-mass control. For SO(2) rotations, rotating values too makes residual far contributions combine incoherently, extending the range. Controlled experiments support these predictions: random accumulated rotations substantially improve extrapolation over RoPE, learned token-dependent rotations maintain near-training-length perplexity far beyond the training context, and rotating values helps over queries and keys alone. Rotation-only models still degrade at extreme lengths, while ALiBi stays length-stable, consistent with the need for far-mass control.
| Comments: | 33 pages, submitted to TMLR |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| MSC classes: | 68T07, 60B20 |
| ACM classes: | I.2.6; G.3; G.1.3 |
| Cite as: | arXiv:2606.24975 [cs.LG] |
| (or arXiv:2606.24975v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2606.24975
arXiv-issued DOI via DataCite
|
Access Paper:
- View PDF
- HTML (experimental)
- TeX Source
Current browse context:
References & Citations
Bibliographic and Citation Tools
Code, Data and Media Associated with this Article
Demos
Recommenders and Search Tools
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
More from arXiv — NLP / Computation & Language
-
Generating in the Limit with Infinitely Many Hallucinations
Jun 30
-
Extracting Knowledge from an Arabic-English Machine-Readable Dictionary Using Information Extraction
Jun 30
-
Developmental Trajectories of Situation Modeling and Mentalizing in Transformer Language Models
Jun 30
-
A French OSCE Dialogue Dataset and Controllable Virtual Patient System for Clinical Training
Jun 30
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.