{
  "abstract": "Diffusion-based generative models typically rely on a sequence of noise perturbations during training and a learned reverse process during generation. Although contraction can occur within the dynamics of existing generative models, conventional sampling procedures do not necessarily formulate generation as an explicitly controlled alternation between expansion and contraction phases. This paper proposes Oscillatory Expansion–Contraction Diffusion (OECD), a conceptual framework in which the generative state alternates between stochastic expansion and learned contraction. During expansion, controlled stochastic perturbations encourage exploration of alternative states. During contraction, a learned operator moves the current state toward regions considered more compatible with the learned data distribution. A switching controller determines the duration, intensity, and order of these phases according to state uncertainty, local stability, estimated data compatibility, and computational cost. Unlike a fixed noise schedule, the proposed mechanism permits state-dependent switching and variable-length phases. We formalize the framework as a stochastic state-transition process, define alternative contraction operators, introduce a training objective, and propose an experimental protocol for image generation. Relevant existing research on contractive diffusion models, nested diffusion, and score-based generation establishes important overlap with the proposed components. Therefore, the primary research question is whether explicitly controlled oscillation between expansion and contraction provides measurable advantages over established diffusion samplers. No experimental results or performance improvements are claimed in this paper.",
  "abstract_shown": true,
  "affiliations": [],
  "article_type": null,
  "arxiv": null,
  "author_total": 1,
  "authors": [
    {
      "name": "Jincheng Zhang",
      "orcid": "0009-0005-1860-0009"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/3rbet5ed9687c.bib",
  "cited_by": 0,
  "doi": "10.5281/zenodo.23615540",
  "fulltext_url": null,
  "id": "3rbet5ed9687c",
  "issns": [],
  "landing_url": "https://zenodo.org/records/23615540",
  "language": null,
  "licence": "CC-BY-4.0",
  "licence_label": "CC BY 4.0",
  "licence_source": "article",
  "licence_url": "https://creativecommons.org/licenses/by/4.0/",
  "metadata_licence": null,
  "native_subjects": [],
  "openalex": null,
  "pdf_url": "https://zenodo.org/api/records/23615540/files/Oscillatory%20Expansion%E2%80%93Contraction%20Diffusion%20A%20Dynamically%20Switching%20Framework%20for%20Image%20Generation.pdf/content",
  "pmcid": null,
  "pmid": null,
  "published": "2026-10-24",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/3rbet5ed9687c.ris",
  "sources": [
    {
      "first_seen": "2026-11-07T12:20:54Z",
      "native_id": "23615540",
      "source": "zenodo",
      "url": "https://zenodo.org/records/23615540"
    }
  ],
  "subjects": [],
  "title": "Oscillatory Expansion–Contraction Diffusion: A Dynamically Switching Framework for Image Generation",
  "title_text": "Oscillatory Expansion–Contraction Diffusion: A Dynamically Switching Framework for Image Generation",
  "type": "preprint",
  "updated_at": "2026-11-07T12:20:54Z",
  "url": "https://research.draconitic.com/paper/3rbet5ed9687c",
  "venue": "Zenodo",
  "versions": []
}