{"slug": "marginal-utility-matrix-factorization-and-the-key-value-kv-cache-a-unified-for", "title": "Marginal utility, matrix factorization, and the Key-Value (KV) cache: a unified information-economic framework for sovereign geo-mining inference", "summary": "A new arXiv paper (2609.20068v1) unifies marginal utility, matrix factorization, and the transformer Key-Value cache under a single allocation rule that retains the top dimensions whose eigenvalue exceeds the shadow price of the binding constraint. The paper reports an 11.2-million-parameter hierarchical classifier, trained in about five minutes on a single GPU, reaching 90.0 per cent level-1 accuracy on a held-out test set from a 973-document uranium-exploration corpus, versus 92.0 per cent for a proprietary model on a fifty-document human audit of the same corpus, at 2.62 ms per card against roughly 2,000 ms for the API. A diagnostic of uniform-density TIES merging exposed a reproducible degenerate mode in which the merged model returned token-identical outputs across five geographically distinct districts while declaring high confidence; re-executing the merge under layer-wise calibrated densities removed that signature on the diagnostic sample, and the full-scale extraction benchmark including LoRA fine-tuning is reported as projected rather than measured.", "body_md": "arXiv:2609.20068v1 Announce Type: new \nAbstract: This paper builds a theoretical bridge between the economic notion of marginal utility and two machine-learning constructs, matrix factorization and the Key--Value cache of transformer language models. The singular value spectrum of a rating matrix is shown to be a diminishing marginal utility schedule for latent factors, the eigenvalue spectrum of the projected covariance operator to be the marginal utility schedule of a model's learned representation, and cache eviction and low-rank cache compression to be instances of constrained utility maximization under a memory budget. The three collapse into a single allocation rule: retain the top dimensions whose eigenvalue exceeds the shadow price of the binding constraint. The framework is applied to the automated extraction of structured information from geo-mining documents, where it motivates a multi-pass inference protocol, a layer-wise TIES model merging procedure, and a selection policy combining extraction quality, localization drift and energy, scalarized with a Conditional Value-at-Risk term on drift. Two empirical contributions are reported. An 11.2-million-parameter hierarchical classifier, trained in about five minutes on a single GPU, reaches 90.0 per cent level-1 accuracy on a held-out test set from a 973-document uranium-exploration corpus, against 92.0 per cent for a proprietary model on a fifty-document human audit of the same corpus, at a latency of 2.62 ms per card against approximately 2,000 ms for the API and at negligible cost. A diagnostic of uniform-density TIES merging exposes a reproducible degenerate mode in which the merged model returns token-identical outputs across five geographically distinct districts while declaring high confidence; re-executing the merge under layer-wise calibrated densities removes that signature on the diagnostic sample. The full-scale extraction benchmark, including LoRA fine-tuning, is reported as projected rather than measured and remains an empirical extension of this work.", "url": "https://wpnews.pro/news/marginal-utility-matrix-factorization-and-the-key-value-kv-cache-a-unified-for", "canonical_source": "https://www.machinebrief.com/news/marginal-utility-matrix-factorization-and-the-key-value-kv-c-l2x1", "published_at": "2026-09-18 04:00:00+00:00", "updated_at": "2026-09-18 04:54:46.205376+00:00", "lang": "en", "topics": ["machine-learning", "large-language-models", "ai-research", "natural-language-processing", "ai-infrastructure"], "entities": ["arXiv", "Key-Value cache", "TIES", "LoRA", "Conditional Value-at-Risk"], "alternates": {"html": "https://wpnews.pro/news/marginal-utility-matrix-factorization-and-the-key-value-kv-cache-a-unified-for", "markdown": "https://wpnews.pro/news/marginal-utility-matrix-factorization-and-the-key-value-kv-cache-a-unified-for.md", "text": "https://wpnews.pro/news/marginal-utility-matrix-factorization-and-the-key-value-kv-cache-a-unified-for.txt", "jsonld": "https://wpnews.pro/news/marginal-utility-matrix-factorization-and-the-key-value-kv-cache-a-unified-for.jsonld"}}