{
  "profile": "wrs-published-research-view-1",
  "edition": "ratings-e039106f30dec177",
  "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS",
  "entry": {
    "id": "hester-et-al-deep-q-demonstrations-2017",
    "title": "DQfD: demonstration pretraining with temporal-difference and margin losses",
    "summary": "Pretrained a deep Q-network on demonstration data using temporal-difference, large-margin action-classification and regularization losses, then mixed preserved demonstrations with self-generated replay during learning.",
    "boundary": "Original v1 uses separate demonstration/agent buffers with a controlled fixed sampling proportion; do not import later prioritized-replay/n-step revisions automatically. Inherits Double DQN and supervised imitation losses. Better initial Atari performance is not a safety certificate for real systems or expert-free learning.",
    "area": "Demonstration-assisted reinforcement learning",
    "topic": "Artificial intelligence",
    "year": 2017,
    "lane": "HUMAN",
    "result": "ALGORITHM",
    "package_id": "hester-et-al-deep-q-demonstrations-2017",
    "target": {
      "record_id": "urn:uuid:855ef364-a631-4692-adba-dc4969c88823",
      "version": 1,
      "content_hash": "sha256:d4aad7ca8d7d56aedaeece228865e571983267e1d05ef34e49d624d747ec831b"
    },
    "humans": [
      "Todd Hester",
      "Matej Vecerik",
      "Olivier Pietquin",
      "Marc Lanctot",
      "Tom Schaul",
      "Bilal Piot",
      "Andrew Sendonaris",
      "Gabriel Dulac-Arnold",
      "Ian Osband",
      "John Agapiou",
      "Joel Z. Leibo",
      "Audrunas Gruslys"
    ],
    "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": [],
    "related": [],
    "subresults": [],
    "scientific_assessment": {
      "source_review": "SOURCE_CURATED",
      "scientific_review": "NOT_REVIEWED_BY_WRS",
      "formal_replay": "NOT_RUN",
      "provider_attestation": "NONE"
    },
    "sources": [
      {
        "id": "cur20261015-arxiv-1704-03732v1",
        "title": "Learning from Demonstrations for Real World Reinforcement Learning",
        "url": "https://arxiv.org/abs/1704.03732v1"
      }
    ],
    "points": "125000",
    "published_grade": {
      "id": "hester-et-al-deep-q-demonstrations-2017",
      "entry_id": "hester-et-al-deep-q-demonstrations-2017",
      "band": "A",
      "modifier": null,
      "target": {
        "record_id": "urn:uuid:855ef364-a631-4692-adba-dc4969c88823",
        "version": 1,
        "content_hash": "sha256:d4aad7ca8d7d56aedaeece228865e571983267e1d05ef34e49d624d747ec831b"
      },
      "what_changed": "Pretrained a deep Q-network on demonstration data using temporal-difference, large-margin action-classification and regularization losses, then mixed preserved demonstrations with self-generated replay during learning.",
      "scope": "Pretrained a deep Q-network on demonstration data using temporal-difference, large-margin action-classification and regularization losses, then mixed preserved demonstrations with self-generated replay during learning. Original v1 uses separate demonstration/agent buffers with a controlled fixed sampling proportion; do not import later prioritized-replay/n-step revisions automatically. Inherits Double DQN and supervised imitation losses. Better initial Atari performance is not a safety certificate for real systems or expert-free learning.",
      "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:855ef364-a631-4692-adba-dc4969c88823",
        "version": 1,
        "content_hash": "sha256:d4aad7ca8d7d56aedaeece228865e571983267e1d05ef34e49d624d747ec831b"
      },
      "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": "Todd Hester",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Learning from Demonstrations for Real World Reinforcement Learning",
            "url": "https://arxiv.org/abs/1704.03732v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Matej Vecerik",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Learning from Demonstrations for Real World Reinforcement Learning",
            "url": "https://arxiv.org/abs/1704.03732v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Olivier Pietquin",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Learning from Demonstrations for Real World Reinforcement Learning",
            "url": "https://arxiv.org/abs/1704.03732v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Marc Lanctot",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Learning from Demonstrations for Real World Reinforcement Learning",
            "url": "https://arxiv.org/abs/1704.03732v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Tom Schaul",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Learning from Demonstrations for Real World Reinforcement Learning",
            "url": "https://arxiv.org/abs/1704.03732v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Bilal Piot",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Learning from Demonstrations for Real World Reinforcement Learning",
            "url": "https://arxiv.org/abs/1704.03732v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Andrew Sendonaris",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Learning from Demonstrations for Real World Reinforcement Learning",
            "url": "https://arxiv.org/abs/1704.03732v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Gabriel Dulac-Arnold",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Learning from Demonstrations for Real World Reinforcement Learning",
            "url": "https://arxiv.org/abs/1704.03732v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Ian Osband",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Learning from Demonstrations for Real World Reinforcement Learning",
            "url": "https://arxiv.org/abs/1704.03732v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "John Agapiou",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Learning from Demonstrations for Real World Reinforcement Learning",
            "url": "https://arxiv.org/abs/1704.03732v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Joel Z. Leibo",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Learning from Demonstrations for Real World Reinforcement Learning",
            "url": "https://arxiv.org/abs/1704.03732v1"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Audrunas Gruslys",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Learning from Demonstrations for Real World Reinforcement Learning",
            "url": "https://arxiv.org/abs/1704.03732v1"
          }
        }
      ],
      "evidence": [
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-0",
          "kind": "source_reference",
          "source": {
            "citation": "Learning from Demonstrations for Real World Reinforcement Learning",
            "url": "https://arxiv.org/abs/1704.03732v1"
          }
        }
      ],
      "origin": "HUMAN_LEGACY",
      "record_profile": "wrs-local-research-notes-0.1",
      "record_type": "research_note",
      "rrp_version": "0.1",
      "scope": {
        "description": "Original v1 uses separate demonstration/agent buffers with a controlled fixed sampling proportion; do not import later prioritized-replay/n-step revisions automatically. Inherits Double DQN and supervised imitation losses. Better initial Atari performance is not a safety certificate for real systems or expert-free learning.",
        "domain": "Demonstration-assisted reinforcement learning"
      },
      "sources": [
        {
          "citation": "Learning from Demonstrations for Real World Reinforcement Learning",
          "url": "https://arxiv.org/abs/1704.03732v1"
        }
      ],
      "statement": "# DQfD: demonstration pretraining with temporal-difference and margin losses\n\nPretrained a deep Q-network on demonstration data using temporal-difference, large-margin action-classification and regularization losses, then mixed preserved demonstrations with self-generated replay during learning.\n\n## Scientific scope\n\nOriginal v1 uses separate demonstration/agent buffers with a controlled fixed sampling proportion; do not import later prioritized-replay/n-step revisions automatically. Inherits Double DQN and supervised imitation losses. Better initial Atari performance is not a safety certificate for real systems or expert-free learning.\n\n## Reading status\n\nThis is a source-linked WRS catalogue note. A full explanatory guide and independent scientific reconstruction have not been completed for this entry. Read the identified original work for its complete narrative and arguments.\n\n## People and systems\n\nTodd Hester; Matej Vecerik; Olivier Pietquin; Marc Lanctot; Tom Schaul; Bilal Piot; Andrew Sendonaris; Gabriel Dulac-Arnold; Ian Osband; John Agapiou; Joel Z. Leibo; Audrunas Gruslys\n\nSource-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.\n\n\n\n## Original sources\n\n- [Learning from Demonstrations for Real World Reinforcement Learning](https://arxiv.org/abs/1704.03732v1)\n",
      "summary": "Pretrained a deep Q-network on demonstration data using temporal-difference, large-margin action-classification and regularization losses, then mixed preserved demonstrations with self-generated replay during learning.",
      "title": "DQfD: demonstration pretraining with temporal-difference and margin losses"
    },
    "record_id": "urn:uuid:855ef364-a631-4692-adba-dc4969c88823",
    "version": 1,
    "source_ref": {
      "record_id": "urn:uuid:855ef364-a631-4692-adba-dc4969c88823",
      "version": 1,
      "content_hash": "sha256:d4aad7ca8d7d56aedaeece228865e571983267e1d05ef34e49d624d747ec831b"
    },
    "export_view": "wrs-public-editorial-view-1",
    "public_view_hash": "sha256:e2394039bf8ab13b6111f2c5ba12acb0000c21d6b2b1a75a0df91834b684f65b"
  },
  "target": {
    "record_id": "urn:uuid:855ef364-a631-4692-adba-dc4969c88823",
    "version": 1,
    "content_hash": "sha256:d4aad7ca8d7d56aedaeece228865e571983267e1d05ef34e49d624d747ec831b"
  },
  "published_grade": {
    "profile": "wrs-published-grade-1",
    "edition": "2026-10-28.1",
    "id": "hester-et-al-deep-q-demonstrations-2017",
    "entry_id": "hester-et-al-deep-q-demonstrations-2017",
    "revision": 1,
    "title": "DQfD: demonstration pretraining with temporal-difference and margin losses",
    "context": "Demonstration-assisted reinforcement learning",
    "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": "Pretrained a deep Q-network on demonstration data using temporal-difference, large-margin action-classification and regularization losses, then mixed preserved demonstrations with self-generated replay during learning.",
    "scope": "Pretrained a deep Q-network on demonstration data using temporal-difference, large-margin action-classification and regularization losses, then mixed preserved demonstrations with self-generated replay during learning. Original v1 uses separate demonstration/agent buffers with a controlled fixed sampling proportion; do not import later prioritized-replay/n-step revisions automatically. Inherits Double DQN and supervised imitation losses. Better initial Atari performance is not a safety certificate for real systems or expert-free learning.",
    "evidence_summary": "The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Original v1 uses separate demonstration/agent buffers with a controlled fixed sampling proportion; do not import later prioritized-replay/n-step revisions automatically. Inherits Double DQN and supervised imitation losses. Better initial Atari performance is not a safety certificate for real systems or expert-free learning.",
    "references": [
      {
        "title": "Learning from Demonstrations for Real World Reinforcement Learning",
        "url": "https://arxiv.org/abs/1704.03732v1",
        "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:855ef364-a631-4692-adba-dc4969c88823",
      "version": 1,
      "content_hash": "sha256:d4aad7ca8d7d56aedaeece228865e571983267e1d05ef34e49d624d747ec831b"
    },
    "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"
}