{
  "abstract": "We study in-context binary classification in a simple attention model trained to minimise a logistic loss. Each prompt has a latent logistic teacher, and the learner must infer its direction from Bernoulli-labeled context examples. We derive an asymptotic prediction for the test logit in a high-dimensional regime with context length proportional to token dimension, and total training prompts proportional to dimension squared. Our formula cleanly separates calibration, test-time context noise, and non-isotropic parameter-estimation noise as interpretable effects on in-context performance. Our analysis also identifies the ridgeless separability regime and addresses the effects of context length, pretraining-set size, regularisation, and sampling temperature on in-context classification performance. The theory's predictions closely agree with finite-dimensional simulations and fully trained linear-attention models, while experiments with standard softmax attention show qualitatively related behaviour beyond the regime assumed by our analysis.",
  "abstract_shown": true,
  "affiliations": [],
  "article_type": null,
  "arxiv": "2610.17061",
  "author_total": 5,
  "authors": [
    {
      "name": "Arman Rysmakhanov"
    },
    {
      "name": "Mary Letey"
    },
    {
      "name": "Yue M. Lu"
    },
    {
      "name": "Jacob Zavatone-Veth"
    },
    {
      "name": "Cengiz Pehlevan"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/f3s137w51prnr.bib",
  "cited_by": 0,
  "doi": "10.48550/arxiv.2610.17061",
  "fulltext_url": null,
  "id": "f3s137w51prnr",
  "issns": [],
  "landing_url": "https://arxiv.org/abs/2610.17061",
  "language": null,
  "licence": "CC-BY-4.0",
  "licence_label": "CC BY 4.0",
  "licence_source": "article",
  "licence_url": "http://creativecommons.org/licenses/by/4.0/",
  "metadata_licence": "CC0-1.0",
  "native_subjects": [
    [
      "arxiv",
      "cs.LG"
    ],
    [
      "arxiv",
      "stat.ML"
    ],
    [
      "openalex",
      "1702|Domain Adaptation and Few-Shot Learning"
    ]
  ],
  "openalex": "W7221770880",
  "pdf_url": "https://arxiv.org/pdf/2610.17061",
  "pmcid": null,
  "pmid": null,
  "published": "2026-10-12",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/f3s137w51prnr.ris",
  "sources": [
    {
      "first_seen": "2026-10-30T12:29:26Z",
      "native_id": "2610.17061",
      "source": "arxiv",
      "url": "https://arxiv.org/abs/2610.17061"
    }
  ],
  "subjects": [
    {
      "field": "mathematics",
      "origin": "arxiv:stat.ML",
      "subfield": "statistics"
    }
  ],
  "title": "In-context classification: Calibration and noise in linear attention with softmax readout",
  "title_text": "In-context classification: Calibration and noise in linear attention with softmax readout",
  "type": "preprint",
  "updated_at": "2026-10-30T12:29:26Z",
  "url": "https://research.draconitic.com/paper/f3s137w51prnr",
  "venue": "arXiv",
  "versions": []
}