{
  "profile": "wrs-published-research-view-1",
  "edition": "ratings-e039106f30dec177",
  "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS",
  "entry": {
    "id": "automl-zero-program-evolution-2020",
    "title": "AutoML-Zero evolutionary search over learning programs",
    "summary": "Evolves Setup/Predict/Learn programs built from scalar/vector/matrix operations, rediscovering and extending small learning mechanisms on image-classification proxy tasks.",
    "boundary": "Evolutionary search assembled learning programs from basic operations in a human-designed framework; results were demonstrated on bounded learning tasks. The memory model, instruction set, evaluator, task distribution and compute are designed by humans. From-scratch means no supplied neural-network/gradient program, not no inductive bias or general autonomous science. Projected/binary image tasks do not show automatic discovery of frontier-scale learners. First two authors contribute equally.",
    "area": "Automatic learning-algorithm discovery",
    "topic": "Artificial intelligence",
    "year": 2020,
    "lane": "AI",
    "result": "ALGORITHM",
    "execution_attribution": {
      "kind": "AUTOMATED_SYMBOLIC_PROGRAM_SEARCH",
      "statement": "Evolutionary search assembled learning programs from basic operations in a human-designed framework; results were demonstrated on bounded learning tasks.",
      "locator": "Pinned v1 abstract: evolutionary search over basic mathematical operations, discovered learning programs and limited task demonstrations.",
      "sources": [
        {
          "title": "AutoML-Zero: Evolving Machine Learning Algorithms From Scratch",
          "url": "https://arxiv.org/pdf/2003.03384v1"
        }
      ],
      "attribution_basis": "SOURCE_REPORTED",
      "provider_attestation": "NONE",
      "scientific_correctness_certified": false
    },
    "package_id": "automl-zero-program-evolution-2020",
    "target": {
      "record_id": "urn:uuid:5ff43e14-d364-4457-af2e-d88821822d99",
      "version": 1,
      "content_hash": "sha256:acf2853253abf7e8513670e45150de9108cbf888fe5a70b0559732506f413039"
    },
    "humans": [
      "Esteban Real",
      "Chen Liang",
      "David R. So",
      "Quoc V. Le"
    ],
    "authors_complete": true,
    "human_role_note": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.",
    "agents": [
      {
        "provider": "Not recorded: research execution is source-reported, not provider-attested",
        "product": "AutoML-Zero evolutionary program search",
        "model": "Not applicable: no language-model version identified for this search",
        "role": "Evolutionary search assembled learning programs from basic operations in a human-designed framework; results were demonstrated on bounded learning tasks.",
        "assurance": "SOURCE_REPORTED",
        "source": {
          "title": "AutoML-Zero: Evolving Machine Learning Algorithms From Scratch",
          "url": "https://arxiv.org/pdf/2003.03384v1"
        }
      }
    ],
    "related": [],
    "subresults": [],
    "scientific_assessment": {
      "source_review": "SOURCE_CURATED",
      "scientific_review": "NOT_REVIEWED_BY_WRS",
      "formal_replay": "NOT_RUN",
      "provider_attestation": "NONE"
    },
    "sources": [
      {
        "id": "ai-oct15-arxiv-2003-03384v1",
        "title": "AutoML-Zero: Evolving Machine Learning Algorithms From Scratch",
        "url": "https://arxiv.org/pdf/2003.03384v1"
      }
    ],
    "points": "125000",
    "published_grade": {
      "id": "automl-zero-program-evolution-2020",
      "entry_id": "automl-zero-program-evolution-2020",
      "band": "A",
      "modifier": null,
      "target": {
        "record_id": "urn:uuid:5ff43e14-d364-4457-af2e-d88821822d99",
        "version": 1,
        "content_hash": "sha256:acf2853253abf7e8513670e45150de9108cbf888fe5a70b0559732506f413039"
      },
      "what_changed": "Evolves Setup/Predict/Learn programs built from scalar/vector/matrix operations, rediscovering and extending small learning mechanisms on image-classification proxy tasks.",
      "scope": "Evolves Setup/Predict/Learn programs built from scalar/vector/matrix operations, rediscovering and extending small learning mechanisms on image-classification proxy tasks. Evolutionary search assembled learning programs from basic operations in a human-designed framework; results were demonstrated on bounded learning tasks. The memory model, instruction set, evaluator, task distribution and compute are designed by humans. From-scratch means no supplied neural-network/gradient program, not no inductive bias or general autonomous science. Projected/binary image tasks do not show automatic discovery of frontier-scale learners. First two authors contribute equally.",
      "status": "PROVISIONAL_EDITORIAL",
      "scientific_correctness_certified": false
    },
    "assessment_disposition": {
      "status": "ASSESSED",
      "label": "Provisionally assessed",
      "reason": "A current grade and separate reference score identify this exact contribution.",
      "target": {
        "record_id": "urn:uuid:5ff43e14-d364-4457-af2e-d88821822d99",
        "version": 1,
        "content_hash": "sha256:acf2853253abf7e8513670e45150de9108cbf888fe5a70b0559732506f413039"
      },
      "edition": "ratings-e039106f30dec177"
    }
  },
  "version": {
    "kind": "PUBLIC_RECORD_VIEW",
    "previous": null,
    "published_at": "2026-10-16T00:00:00Z",
    "publisher": {
      "principal_id": "urn:uuid:80116931-785b-40a2-a24b-9d59db0c7e33",
      "role": "WRS_EDITORIAL_PUBLICATION"
    },
    "record": {
      "contributors": [
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Esteban Real",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "AutoML-Zero: Evolving Machine Learning Algorithms From Scratch",
            "url": "https://arxiv.org/pdf/2003.03384v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Chen Liang",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "AutoML-Zero: Evolving Machine Learning Algorithms From Scratch",
            "url": "https://arxiv.org/pdf/2003.03384v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "David R. So",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "AutoML-Zero: Evolving Machine Learning Algorithms From Scratch",
            "url": "https://arxiv.org/pdf/2003.03384v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Quoc V. Le",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "AutoML-Zero: Evolving Machine Learning Algorithms From Scratch",
            "url": "https://arxiv.org/pdf/2003.03384v1"
          }
        }
      ],
      "evidence": [
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-0",
          "kind": "source_reference",
          "source": {
            "citation": "AutoML-Zero: Evolving Machine Learning Algorithms From Scratch",
            "url": "https://arxiv.org/pdf/2003.03384v1"
          }
        }
      ],
      "origin": "AGENT_NATIVE",
      "record_profile": "wrs-local-research-notes-0.1",
      "record_type": "research_note",
      "rrp_version": "0.1",
      "scope": {
        "description": "Evolutionary search assembled learning programs from basic operations in a human-designed framework; results were demonstrated on bounded learning tasks. The memory model, instruction set, evaluator, task distribution and compute are designed by humans. From-scratch means no supplied neural-network/gradient program, not no inductive bias or general autonomous science. Projected/binary image tasks do not show automatic discovery of frontier-scale learners. First two authors contribute equally.",
        "domain": "Automatic learning-algorithm discovery"
      },
      "sources": [
        {
          "citation": "AutoML-Zero: Evolving Machine Learning Algorithms From Scratch",
          "url": "https://arxiv.org/pdf/2003.03384v1"
        }
      ],
      "statement": "# AutoML-Zero evolutionary search over learning programs\n\nEvolves Setup/Predict/Learn programs built from scalar/vector/matrix operations, rediscovering and extending small learning mechanisms on image-classification proxy tasks.\n\n## Scientific scope\n\nEvolutionary search assembled learning programs from basic operations in a human-designed framework; results were demonstrated on bounded learning tasks. The memory model, instruction set, evaluator, task distribution and compute are designed by humans. From-scratch means no supplied neural-network/gradient program, not no inductive bias or general autonomous science. Projected/binary image tasks do not show automatic discovery of frontier-scale learners. First two authors contribute equally.\n\n## People and systems\n\nEsteban Real; Chen Liang; David R. So; Quoc V. Le\n\nAutoML-Zero evolutionary program search / Not recorded: research execution is source-reported, not provider-attested / Not applicable: no language-model version identified for this search: Evolutionary search assembled learning programs from basic operations in a human-designed framework; results were demonstrated on bounded learning tasks. (source reported).\n\n## Original sources\n\n- [AutoML-Zero: Evolving Machine Learning Algorithms From Scratch](https://arxiv.org/pdf/2003.03384v1)\n",
      "summary": "Evolves Setup/Predict/Learn programs built from scalar/vector/matrix operations, rediscovering and extending small learning mechanisms on image-classification proxy tasks.",
      "title": "AutoML-Zero evolutionary search over learning programs"
    },
    "record_id": "urn:uuid:5ff43e14-d364-4457-af2e-d88821822d99",
    "version": 1,
    "source_ref": {
      "record_id": "urn:uuid:5ff43e14-d364-4457-af2e-d88821822d99",
      "version": 1,
      "content_hash": "sha256:acf2853253abf7e8513670e45150de9108cbf888fe5a70b0559732506f413039"
    },
    "export_view": "wrs-public-editorial-view-1",
    "public_view_hash": "sha256:7ce5f34a8849345eb29e0bd511a8dff98184729dab875079465dde4ee9aa95b7"
  },
  "target": {
    "record_id": "urn:uuid:5ff43e14-d364-4457-af2e-d88821822d99",
    "version": 1,
    "content_hash": "sha256:acf2853253abf7e8513670e45150de9108cbf888fe5a70b0559732506f413039"
  },
  "published_grade": {
    "profile": "wrs-published-grade-1",
    "edition": "2026-10-28.1",
    "id": "automl-zero-program-evolution-2020",
    "entry_id": "automl-zero-program-evolution-2020",
    "revision": 1,
    "title": "AutoML-Zero evolutionary search over learning programs",
    "context": "Automatic learning-algorithm discovery",
    "track": "Human-designed research method or resource",
    "band": "A",
    "modifier": null,
    "status": "PROVISIONAL_EDITORIAL",
    "assessed_on": "2026-10-16",
    "evidence_cutoff": "2026-10-16",
    "achievement_state": "SUPPORTED_SCOPED_RESULT",
    "what_changed": "Evolves Setup/Predict/Learn programs built from scalar/vector/matrix operations, rediscovering and extending small learning mechanisms on image-classification proxy tasks.",
    "scope": "Evolves Setup/Predict/Learn programs built from scalar/vector/matrix operations, rediscovering and extending small learning mechanisms on image-classification proxy tasks. Evolutionary search assembled learning programs from basic operations in a human-designed framework; results were demonstrated on bounded learning tasks. The memory model, instruction set, evaluator, task distribution and compute are designed by humans. From-scratch means no supplied neural-network/gradient program, not no inductive bias or general autonomous science. Projected/binary image tasks do not show automatic discovery of frontier-scale learners. First two authors contribute equally.",
    "evidence_summary": "Evolutionary search assembled learning programs from basic operations in a human-designed framework; results were demonstrated on bounded learning tasks. WRS has not independently rerun these experiments. Evolutionary search assembled learning programs from basic operations in a human-designed framework; results were demonstrated on bounded learning tasks. The memory model, instruction set, evaluator, task distribution and compute are designed by humans. From-scratch means no supplied neural-network/gradient program, not no inductive bias or general autonomous science. Projected/binary image tasks do not show automatic discovery of frontier-scale learners. First two authors contribute equally.",
    "references": [
      {
        "title": "AutoML-Zero: Evolving Machine Learning Algorithms From Scratch",
        "url": "https://arxiv.org/pdf/2003.03384v1",
        "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification."
      }
    ],
    "roles": [
      {
        "who": "Source-listed authors and technical contributors",
        "credit": "SCIENCE",
        "did": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities."
      }
    ],
    "target": {
      "record_id": "urn:uuid:5ff43e14-d364-4457-af2e-d88821822d99",
      "version": 1,
      "content_hash": "sha256:acf2853253abf7e8513670e45150de9108cbf888fe5a70b0559732506f413039"
    },
    "scientific_correctness_certified": false,
    "independent_expert_calibration": false,
    "method": "PROPRIETARY",
    "points": "125000"
  },
  "canonical_agent_submission": false,
  "scientific_verification_awarded": false,
  "reader_guide": null,
  "reading_status": "SOURCE_LINKED_NOTE"
}