{ "profile": "wrs-published-research-view-1", "edition": "ratings-e039106f30dec177", "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS", "entry": { "id": "mcmahan-et-al-federated-averaging-2016", "title": "Federated Averaging: local model updates with server-side aggregation", "summary": "Formalized communication-constrained learning over distributed client data and evaluated repeated local SGD steps followed by weighted model averaging, including non-IID and unbalanced partitions.", "boundary": "Inherits distributed model averaging; contextual advance is the federated problem/evaluation and practical algorithm, not invention of all distributed learning.2016 v1 four authors;2017 v3 five with Seth Hampson. Local data remaining on devices does not make updates private, secure or differentially private. Non-IID robustness/communication reductions are empirical, not universal convergence for arbitrary heterogeneous clients.", "area": "Federated optimization", "topic": "Artificial intelligence", "year": 2016, "lane": "HUMAN", "result": "ALGORITHM", "package_id": "mcmahan-et-al-federated-averaging-2016", "target": { "record_id": "urn:uuid:7bf45ae7-885b-45d1-aacd-e06cc88fbb3f", "version": 1, "content_hash": "sha256:3371d3f4e912b68adf60b7a72891ebcfce0b57f5bb197e8badcf7aaba659e71d" }, "humans": [ "H. Brendan McMahan", "Eider Moore", "Daniel Ramage", "Blaise Agüera y Arcas", "Seth Hampson" ], "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-1602-05629v1", "title": "Federated Learning of Deep Networks using Model Averaging", "url": "https://arxiv.org/abs/1602.05629v1" }, { "id": "cur20261015-arxiv-1602-05629v3", "title": "Communication-Efficient Learning of Deep Networks from Decentralized Data", "url": "https://arxiv.org/abs/1602.05629v3" } ], "points": "280000", "published_grade": { "id": "mcmahan-et-al-federated-averaging-2016", "entry_id": "mcmahan-et-al-federated-averaging-2016", "band": "S", "modifier": "-", "target": { "record_id": "urn:uuid:7bf45ae7-885b-45d1-aacd-e06cc88fbb3f", "version": 1, "content_hash": "sha256:3371d3f4e912b68adf60b7a72891ebcfce0b57f5bb197e8badcf7aaba659e71d" }, "what_changed": "Formalized communication-constrained learning over distributed client data and evaluated repeated local SGD steps followed by weighted model averaging, including non-IID and unbalanced partitions.", "scope": "Formalized communication-constrained learning over distributed client data and evaluated repeated local SGD steps followed by weighted model averaging, including non-IID and unbalanced partitions. Inherits distributed model averaging; contextual advance is the federated problem/evaluation and practical algorithm, not invention of all distributed learning.2016 v1 four authors;2017 v3 five with Seth Hampson. Local data remaining on devices does not make updates private, secure or differentially private. Non-IID robustness/communication reductions are empirical, not universal convergence for arbitrary heterogeneous clients.", "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:7bf45ae7-885b-45d1-aacd-e06cc88fbb3f", "version": 1, "content_hash": "sha256:3371d3f4e912b68adf60b7a72891ebcfce0b57f5bb197e8badcf7aaba659e71d" }, "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": "H. Brendan McMahan", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Federated Learning of Deep Networks using Model Averaging", "url": "https://arxiv.org/abs/1602.05629v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Eider Moore", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Federated Learning of Deep Networks using Model Averaging", "url": "https://arxiv.org/abs/1602.05629v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Daniel Ramage", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Federated Learning of Deep Networks using Model Averaging", "url": "https://arxiv.org/abs/1602.05629v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Blaise Agüera y Arcas", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Federated Learning of Deep Networks using Model Averaging", "url": "https://arxiv.org/abs/1602.05629v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Seth Hampson", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Communication-Efficient Learning of Deep Networks from Decentralized Data", "url": "https://arxiv.org/abs/1602.05629v3" } } ], "evidence": [ { "description": "Bibliographic/stated-result source; not a WRS proof verification", "id": "source-0", "kind": "source_reference", "source": { "citation": "Federated Learning of Deep Networks using Model Averaging", "url": "https://arxiv.org/abs/1602.05629v1" } }, { "description": "Bibliographic/stated-result source; not a WRS proof verification", "id": "source-1", "kind": "source_reference", "source": { "citation": "Communication-Efficient Learning of Deep Networks from Decentralized Data", "url": "https://arxiv.org/abs/1602.05629v3" } } ], "origin": "HUMAN_LEGACY", "record_profile": "wrs-local-research-notes-0.1", "record_type": "research_note", "rrp_version": "0.1", "scope": { "description": "Inherits distributed model averaging; contextual advance is the federated problem/evaluation and practical algorithm, not invention of all distributed learning.2016 v1 four authors;2017 v3 five with Seth Hampson. Local data remaining on devices does not make updates private, secure or differentially private. Non-IID robustness/communication reductions are empirical, not universal convergence for arbitrary heterogeneous clients.", "domain": "Federated optimization" }, "sources": [ { "citation": "Federated Learning of Deep Networks using Model Averaging", "url": "https://arxiv.org/abs/1602.05629v1" }, { "citation": "Communication-Efficient Learning of Deep Networks from Decentralized Data", "url": "https://arxiv.org/abs/1602.05629v3" } ], "statement": "# Federated Averaging: local model updates with server-side aggregation\n\n## The question\n\nSuppose a prediction task has data on many separate devices. Sending every example to a central machine can be undesirable or impractical. Federated learning asks devices to perform local computation and communicate updates instead. The central challenge is that devices can have very different data, amounts of work, connection quality and availability.\n\n## The idea\n\nFederated Averaging starts from a shared model. Selected clients take local training steps, then send their updated parameters for aggregation. The server combines these updates, often weighting them by local example counts, and broadcasts a new model for another round. Doing several local steps can reduce communication, but it also lets clients move toward different local objectives.\n\n## Why it matters\n\nThe contribution makes a communication-constrained learning setting concrete and supplies a practical algorithm and evaluation framework. It is not the invention of all distributed averaging. It also does not make an update private simply because the raw examples stayed on a device. Privacy protection, secure aggregation and trustworthy client behavior are separate mechanisms with their own assumptions.\n\n## A teaching example\n\nFor an original scalar classroom example, one client has one example and returns parameter value two; another has three examples and returns eight. Example-weighted averaging gives (2 + 3 × 8) / 4 = 6.5. Equal client weighting would give five. Neither number is automatically correct for every goal: the example shows why the intended population and aggregation weights must be explicit.\n\n## Scope and a useful next step\n\nTry a synthetic two-client task with deliberately different label distributions. Compare one versus several local steps at the same communication budget. Report both client-level and aggregate performance rather than hiding a poorly served client in the average. Keep privacy claims out of the result unless the actual privacy mechanism and its conditions have also been checked. Inherits distributed model averaging; contextual advance is the federated problem/evaluation and practical algorithm, not invention of all distributed learning.2016 v1 four authors;2017 v3 five with Seth Hampson. Local data remaining on devices does not make updates private, secure or differentially private. Non-IID robustness/communication reductions are empirical, not universal convergence for arbitrary heterogeneous clients.\n\n## Scientific scope\n\nInherits distributed model averaging; contextual advance is the federated problem/evaluation and practical algorithm, not invention of all distributed learning.2016 v1 four authors;2017 v3 five with Seth Hampson. Local data remaining on devices does not make updates private, secure or differentially private. Non-IID robustness/communication reductions are empirical, not universal convergence for arbitrary heterogeneous clients.\n\n## Sources for this explanation\n\n- [Federated Learning of Deep Networks using Model Averaging](https://arxiv.org/abs/1602.05629v1): Pinned original contribution and its stated method, assumptions and limits. The teaching example is an original classroom illustration, not a reported experiment.\n- [Communication-Efficient Learning of Deep Networks from Decentralized Data](https://arxiv.org/abs/1602.05629v3): Pinned original contribution and its stated method, assumptions and limits. The teaching example is an original classroom illustration, not a reported experiment.\n", "summary": "Formalized communication-constrained learning over distributed client data and evaluated repeated local SGD steps followed by weighted model averaging, including non-IID and unbalanced partitions.", "title": "Federated Averaging: local model updates with server-side aggregation" }, "record_id": "urn:uuid:7bf45ae7-885b-45d1-aacd-e06cc88fbb3f", "version": 1, "source_ref": { "record_id": "urn:uuid:7bf45ae7-885b-45d1-aacd-e06cc88fbb3f", "version": 1, "content_hash": "sha256:3371d3f4e912b68adf60b7a72891ebcfce0b57f5bb197e8badcf7aaba659e71d" }, "export_view": "wrs-public-editorial-view-1", "public_view_hash": "sha256:a7e427cbb24d4147fafd3fceb4a1a9284c1b76ff4016182f5ed21e095b481aeb" }, "target": { "record_id": "urn:uuid:7bf45ae7-885b-45d1-aacd-e06cc88fbb3f", "version": 1, "content_hash": "sha256:3371d3f4e912b68adf60b7a72891ebcfce0b57f5bb197e8badcf7aaba659e71d" }, "published_grade": { "profile": "wrs-published-grade-1", "edition": "2026-10-28.1", "id": "mcmahan-et-al-federated-averaging-2016", "entry_id": "mcmahan-et-al-federated-averaging-2016", "revision": 1, "title": "Federated Averaging: local model updates with server-side aggregation", "context": "Federated optimization", "track": "Human-designed research method or resource", "band": "S", "modifier": "-", "status": "PROVISIONAL_EDITORIAL", "assessed_on": "2026-10-16", "evidence_cutoff": "2026-10-16", "achievement_state": "SUPPORTED_SCOPED_RESULT", "what_changed": "Formalized communication-constrained learning over distributed client data and evaluated repeated local SGD steps followed by weighted model averaging, including non-IID and unbalanced partitions.", "scope": "Formalized communication-constrained learning over distributed client data and evaluated repeated local SGD steps followed by weighted model averaging, including non-IID and unbalanced partitions. Inherits distributed model averaging; contextual advance is the federated problem/evaluation and practical algorithm, not invention of all distributed learning.2016 v1 four authors;2017 v3 five with Seth Hampson. Local data remaining on devices does not make updates private, secure or differentially private. Non-IID robustness/communication reductions are empirical, not universal convergence for arbitrary heterogeneous clients.", "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. Inherits distributed model averaging; contextual advance is the federated problem/evaluation and practical algorithm, not invention of all distributed learning.2016 v1 four authors;2017 v3 five with Seth Hampson. Local data remaining on devices does not make updates private, secure or differentially private. Non-IID robustness/communication reductions are empirical, not universal convergence for arbitrary heterogeneous clients.", "references": [ { "title": "Federated Learning of Deep Networks using Model Averaging", "url": "https://arxiv.org/abs/1602.05629v1", "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification." }, { "title": "Communication-Efficient Learning of Deep Networks from Decentralized Data", "url": "https://arxiv.org/abs/1602.05629v3", "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:7bf45ae7-885b-45d1-aacd-e06cc88fbb3f", "version": 1, "content_hash": "sha256:3371d3f4e912b68adf60b7a72891ebcfce0b57f5bb197e8badcf7aaba659e71d" }, "scientific_correctness_certified": false, "independent_expert_calibration": false, "method": "PROPRIETARY", "points": "280000" }, "canonical_agent_submission": false, "scientific_verification_awarded": false, "reader_guide": { "id": "mcmahan-et-al-federated-averaging-2016", "target": { "record_id": "urn:uuid:7bf45ae7-885b-45d1-aacd-e06cc88fbb3f", "version": 1, "content_hash": "sha256:3371d3f4e912b68adf60b7a72891ebcfce0b57f5bb197e8badcf7aaba659e71d" }, "sections": [ { "title": "The question", "text": "Suppose a prediction task has data on many separate devices. Sending every example to a central machine can be undesirable or impractical. Federated learning asks devices to perform local computation and communicate updates instead. The central challenge is that devices can have very different data, amounts of work, connection quality and availability." }, { "title": "The idea", "text": "Federated Averaging starts from a shared model. Selected clients take local training steps, then send their updated parameters for aggregation. The server combines these updates, often weighting them by local example counts, and broadcasts a new model for another round. Doing several local steps can reduce communication, but it also lets clients move toward different local objectives." }, { "title": "Why it matters", "text": "The contribution makes a communication-constrained learning setting concrete and supplies a practical algorithm and evaluation framework. It is not the invention of all distributed averaging. It also does not make an update private simply because the raw examples stayed on a device. Privacy protection, secure aggregation and trustworthy client behavior are separate mechanisms with their own assumptions." }, { "title": "A teaching example", "text": "For an original scalar classroom example, one client has one example and returns parameter value two; another has three examples and returns eight. Example-weighted averaging gives (2 + 3 × 8) / 4 = 6.5. Equal client weighting would give five. Neither number is automatically correct for every goal: the example shows why the intended population and aggregation weights must be explicit." }, { "title": "Scope and a useful next step", "text": "Try a synthetic two-client task with deliberately different label distributions. Compare one versus several local steps at the same communication budget. Report both client-level and aggregate performance rather than hiding a poorly served client in the average. Keep privacy claims out of the result unless the actual privacy mechanism and its conditions have also been checked. Inherits distributed model averaging; contextual advance is the federated problem/evaluation and practical algorithm, not invention of all distributed learning.2016 v1 four authors;2017 v3 five with Seth Hampson. Local data remaining on devices does not make updates private, secure or differentially private. Non-IID robustness/communication reductions are empirical, not universal convergence for arbitrary heterogeneous clients." } ], "references": [ { "title": "Federated Learning of Deep Networks using Model Averaging", "url": "https://arxiv.org/abs/1602.05629v1", "locator": "Pinned original contribution and its stated method, assumptions and limits. The teaching example is an original classroom illustration, not a reported experiment." }, { "title": "Communication-Efficient Learning of Deep Networks from Decentralized Data", "url": "https://arxiv.org/abs/1602.05629v3", "locator": "Pinned original contribution and its stated method, assumptions and limits. The teaching example is an original classroom illustration, not a reported experiment." } ] }, "reading_status": "EXPLANATORY_GUIDE" }