AI paper index

The Same Symmetry Lens Reads Different Structure in Each Domain

2026-08-22 · Zenodo (CERN European Organization for Nuclear Research)

One-line summary

An AI research paper on The Same Symmetry Lens Reads Different Structure in Each Domain.

Engineering notes

Engineering notes will be added by the aipentium editorial team.

Chinese explanation / 中文解读

中文解读待补充:本站会优先为大语言模型、生成式AI、ChatGPT相关技术、计算机视觉、深度学习等高价值论文补充中文说明。

Original abstract

The Watabe-Claude Method (WCM) embeds a scalar time series into 4D via a Takens window and reads theoccupancy of the 24-cell regular polytope. The signed-permutation group B4 (order 384) decomposes that24-dimensional occupancy space into six mutually orthogonal isotypic blocks of dimensions 1, 6, 3, 2, 8, 4 (uniform,sign, axispair_a, axispair_b, resid8, resid4). We ask whether these components carry a domain-independentmeaning by running the identical decomposition, unchanged, on real Large-Language-Model (Llama-3.2-3B-Instruct)hidden-state sequences and on real earthquake catalogs (USGS 1986-2026, JMA 1998-2023). They do not. Thesign component opens only when the series oscillates in sign (rate of sign-agreement anti-correlates withsign-component energy across 120 layer/projection conditions, r=-0.37, p=2.7e-5); a smooth series (LLM layerdirection, adjacent cosine 0.906; or an autoregressive generation trajectory) leaves it closed. The ax_a component ismore striking: in static LLM token sequences it fires on local change, whereas in single physical aftershocksequences it fires on flat magnitude persistence -- opposite structure from the same lens. Along the way we rejectedthree tempting unifications, each under an explicit control: ax_a as subject-predicate syntax (dissolved by awindow-position confound); ax_a as a domain-independent sequence-position detector (absent in syntheticaftershocks); and the earthquake signal as a magnitude-quantization artifact (unchanged under jitter). Our centralclaim is a scope result: a symmetry-decomposition component has no single universal referent; what each blockreads is set by the data-generating process, and this is precisely the applicability condition for the method. TextCC-BY-4.0; code AGPL-3.0-or-later. ※ The License for codes is AGPL 3.0-or-later

5.0Engineering value
7.0Research novelty
4.0Business relevance

Links and sources

Need this topic turned into a technical roadmap?

aipentium can prepare a custom AI literature review, code map, dataset map, and B2B technology assessment.

Request B2B AI research

Comments

No comments yet. Be the first to share your thoughts on this paper.
Login or register to leave a comment