{
  "abstract": "Autoencoders are widely used for representation learning, dimensionality reduction, feature extraction, denoising, and generative modeling. Conventional autoencoders learn representations of data, but their architectures, parameters, latent spaces, and encoding–decoding mechanisms are generally designed or optimized at the level of individual models. This paper proposes Meta-Autoencoders (Meta-AEs), a conceptual framework in which autoencoders themselves become objects of representation, reconstruction, generation, and adaptation. Instead of exclusively learning a compressed representation of an input dataset, a Meta-AE learns a representation of an autoencoder or an autoencoder configuration and uses that representation to reconstruct, adapt, or generate an autoencoder suited to a target task.\n\nThe framework distinguishes four complementary operations: autoencoder representation, autoencoder reconstruction, conditional autoencoder generation, and meta-level adaptation. A formal formulation is introduced in which an autoencoder is represented by a structured model description containing its architecture, parameters, latent-space configuration, and optimization characteristics. A meta-encoder maps this description into a compact model representation, while a meta-decoder reconstructs or generates a candidate autoencoder. The proposed objective combines model reconstruction, task-level performance, structural validity, computational cost, and adaptation requirements. A hierarchical extension further permits recursively modeling meta-autoencoders, although such recursion is treated as an optional research direction rather than a demonstrated capability.\n\nThis paper establishes a theoretical framework and an experimental research agenda. It does not claim that the proposed framework has already achieved empirical improvements or that higher-order autoencoder generation is unprecedented. Existing research on hyper-autoencoders, meta-learning, hypernetworks, and neural architecture search provides important foundations and points of comparison. The potential contribution of Meta-AEs lies in investigating a unified model-level framework that can represent, reconstruct, generate, and adapt complete autoencoder configurations under explicit task and resource constraints.",
  "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/26ff77zv4rn67.bib",
  "cited_by": 0,
  "doi": "10.5281/zenodo.23530041",
  "fulltext_url": null,
  "id": "26ff77zv4rn67",
  "issns": [],
  "landing_url": "https://zenodo.org/records/23530041",
  "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/23530041/files/Paper_319_%E5%89%AF%E6%9C%AC8.pdf/content",
  "pmcid": null,
  "pmid": null,
  "published": "2026-10-21",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/26ff77zv4rn67.ris",
  "sources": [
    {
      "first_seen": "2026-11-04T12:21:15Z",
      "native_id": "23530041",
      "source": "zenodo",
      "url": "https://zenodo.org/records/23530041"
    }
  ],
  "subjects": [],
  "title": "Meta-Autoencoders: A Framework for Learning, Reconstructing, and Generating Autoencoder Systems",
  "title_text": "Meta-Autoencoders: A Framework for Learning, Reconstructing, and Generating Autoencoder Systems",
  "type": "preprint",
  "updated_at": "2026-11-04T12:21:15Z",
  "url": "https://research.draconitic.com/paper/26ff77zv4rn67",
  "venue": "Zenodo",
  "versions": []
}