CORTEXA
← Browse
openalexZenodo (CERN European Organization for Nuclear Research)2026-07-23Cited by 0

The correlation geometry of learned representations: a mean-field baseline with a weak anomalous excess, and its cross-domain universality

Rongzhen Dai

The hidden states of large language models carry a power-law correlation along the token axis, C(r) ∝ r^{−α}, whose exponent is stable across model scale and across architectures that share little else. We ask what this apparent “attraction” between nearby positions in representation space actually is. We argue that it is neither a mechanism imposed from outside nor an artefact of a particular training choice, but a near-inevitable consequence of representing structured data compactly — a compression signature. We read the measured exponent as a sum, α = 1/2 + Δ: a mean-field, structureless baseline of one half, plus a weak excess Δ that co-varies with higher-order structure in the data. We then observe that this same “mean-field baseline plus weak anomalous correction” shape is not particular to neural networks. It recurs, with anomalous pieces that others have measured to high precision, in critical phenomena — the anomalous dimension η — and in fully developed turbulence — the intermittency correction δζ₂ — where the anomalous term is universally small, a few percent. We give a sharp criterion separating this structure from a trivial “leading term plus correction”, state carefully what the convergence does and does not establish, and set out the derivation that would be needed to connect the empirical decomposition to an effective-field-theory description of lossy compression.

View free PDFSource page

Related papers

openalexZenodo (CERN European Organization for Nuclear Research)2026-07-24

Before the Model: Why Datasets and Data Representation Define What Machine Learning Can Learn

Jean Franck Loa Rojas

Machine learning systems do not learn reality directly; they learn from the representations preserved in their datasets. This structured narrative review examines how dataset purpose, coverage, integrity, labeling, independence, reproducibility, governance, and continuity determi…

View free PDFSource page
openalexZenodo (CERN European Organization for Nuclear Research)2026-07-24

Convergence-Based Architectures: A Research Monograph on Multi-Perspective Representation Learning through Latent Agreement

Anthony Aseervatham

Convergence-Based Architectures: A New Computational Paradigm for Multi-Perspective Representation Learning This research monograph introduces Convergence-Based Architectures (CBA), a general computational paradigm in which representations emerge through latent agreement among mu…

View free PDFSource page
openalexZenodo (CERN European Organization for Nuclear Research)2026-07-25

Mathematical Framework and Degeneracy Verification of the Five Properties of Matter: Material Space Compression Under the Field Domain Number Equation Constraint

Chao Qin

This paper presents the five-property framework of matter (FC field coherence, FF field frequency, R prime field capacity, PC polymorphic coupling, B4F boundary four-forces), established on the mathematical constraints of the Field Domain Number Equation (FDNE). The framework com…

View free PDFSource page
openalexZenodo (CERN European Organization for Nuclear Research)2026-07-23

Appearance Meets Geometry: Deep Learning for Semantic Segmentation of Archaeological Fortification Masonry

Nils Schnorr, Thomas Leimkühler

This is the preprint for a conference proceedings article to be submitted to the CAA 2025 proceedings. The title of the presentation held by Nils Schnorr during the 2025 conference was "Machine Learning in Archaeological Fortification Studies – A CNN-Based Approach for the Caria…

View free PDFSource page
openalexZenodo (CERN European Organization for Nuclear Research)2026-07-24

Thermodynamic Intelligence: A Fully Analog Neural Network Based on the Information Field with Memristive Learning and Rigorous Mathematical Proofs

Yousefi

Abstract : This paper introduces a novel architecture for intelligent systems, grounded in the natural dynamics of the information field. In this approach, the fundamental concepts of computation and learning are realized not through digital instructions, but through the intrinsi…

View free PDFSource page
openalexZenodo (CERN European Organization for Nuclear Research)2026-07-24

Perfect In-Distribution Accuracy Does Not Imply Learned Physics: Measuring Cross-Family Generalization of Metamaterial Homogenization Surrogates

David Mashiah

Machine-learning surrogates for the effective elastic properties of mechanical metamaterials are almost always trained and evaluated within a single parametric family of unit-cell geometries, with the test set drawn from the same generator as the training set. Under this protocol…

View free PDFSource page