{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/improving-generalization-in-federated","title":"Improving Generalization in Federated Learning by Seeking Flat Minima","arxiv_id":"2203.11834","date":"2022-03-22","proceeding":null,"authors":["Debora Caldarola","Barbara Caputo","Marco Ciccone"],"abstract":"Models trained in federated settings often suffer from degraded performances and fail at generalizing, especially when facing heterogeneous scenarios. In this work, we investigate such behavior through the lens of geometry of the loss and Hessian eigenspectrum, linking the model's lack of generalization capacity to the sharpness of the solution. Motivated by prior studies connecting the sharpness of the loss surface and the generalization gap, we show that i) training clients locally with Sharpness-Aware Minimization (SAM) or its adaptive version (ASAM) and ii) averaging stochastic weights (SWA) on the server-side can substantially improve generalization in Federated Learning and help bridging the gap with centralized models. By seeking parameters in neighborhoods having uniform low loss, the model converges towards flatter minima and its generalization significantly improves in both homogeneous and heterogeneous scenarios. Empirical results demonstrate the effectiveness of those optimizers across a variety of benchmark vision datasets (e.g. CIFAR10/100, Landmarks-User-160k, IDDA) and tasks (large scale classification, semantic segmentation, domain generalization).","url_abs":"https://arxiv.org/abs/2203.11834v3","url_pdf":"https://arxiv.org/pdf/2203.11834v3.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"improving-generalization-in-federated","repo_url":"https://github.com/debcaldarola/fedsam","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"BSD-2-Clause"}}],"tasks":[{"task_slug":"domain-generalization","task_name":"Domain Generalization"},{"task_slug":"federated-learning","task_name":"Federated Learning"},{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"}],"methods":[{"method_slug":"sharpness-aware-minimization","method_name":"Sharpness-Aware Minimization"},{"method_slug":"stochastic-weight-averaging","method_name":"Stochastic Weight Averaging"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-10","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0, 10 clients per round)","model":"FedASAM + SWA","rank_in_archive_order":1,"of":5,"metrics":{"ACC@1-100Clients":"42.64"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-10","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0, 10 clients per round)","model":"FedASAM","rank_in_archive_order":2,"of":5,"metrics":{"ACC@1-100Clients":"39.76"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-10","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0, 10 clients per round)","model":"FedSAM + SWA","rank_in_archive_order":3,"of":5,"metrics":{"ACC@1-100Clients":"39.51"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-10","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0, 10 clients per round)","model":"FedSAM","rank_in_archive_order":4,"of":5,"metrics":{"ACC@1-100Clients":"36.93"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-10","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0, 10 clients per round)","model":"FedAvg","rank_in_archive_order":5,"of":5,"metrics":{"ACC@1-100Clients":"36.74"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-20","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0, 20 clients per round)","model":"FedASAM + SWA","rank_in_archive_order":1,"of":5,"metrics":{"ACC@1-100Clients":"41.62"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-20","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0, 20 clients per round)","model":"FedASAM","rank_in_archive_order":2,"of":5,"metrics":{"ACC@1-100Clients":"40.81"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-20","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0, 20 clients per round)","model":"FedSAM + SWA","rank_in_archive_order":3,"of":5,"metrics":{"ACC@1-100Clients":"39.24"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-20","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0, 20 clients per round)","model":"FedAvg","rank_in_archive_order":4,"of":5,"metrics":{"ACC@1-100Clients":"38.59"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-20","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0, 20 clients per round)","model":"FedSAM","rank_in_archive_order":5,"of":5,"metrics":{"ACC@1-100Clients":"38.56"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0, 5 clients per round)","model":"FedASAM + SWA","rank_in_archive_order":1,"of":5,"metrics":{"ACC@1-100Clients":"42.01"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0, 5 clients per round)","model":"FedSAM + SWA","rank_in_archive_order":2,"of":5,"metrics":{"ACC@1-100Clients":"39.3"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0, 5 clients per round)","model":"FedASAM","rank_in_archive_order":3,"of":5,"metrics":{"ACC@1-100Clients":"36.04"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0, 5 clients per round)","model":"FedSAM","rank_in_archive_order":4,"of":5,"metrics":{"ACC@1-100Clients":"31.04"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0, 5 clients per round)","model":"FedAvg","rank_in_archive_order":5,"of":5,"metrics":{"ACC@1-100Clients":"30.25"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5-10","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0.5, 10 clients per round)","model":"FedASAM + SWA","rank_in_archive_order":1,"of":5,"metrics":{"ACC@1-100Clients":"48.72"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5-10","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0.5, 10 clients per round)","model":"FedSAM + SWA","rank_in_archive_order":2,"of":5,"metrics":{"ACC@1-100Clients":"46.76"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5-10","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0.5, 10 clients per round)","model":"FedASAM","rank_in_archive_order":3,"of":5,"metrics":{"ACC@1-100Clients":"46.58"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5-10","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0.5, 10 clients per round)","model":"FedSAM","rank_in_archive_order":4,"of":5,"metrics":{"ACC@1-100Clients":"44.84"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5-10","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0.5, 10 clients per round)","model":"FedAvg","rank_in_archive_order":5,"of":5,"metrics":{"ACC@1-100Clients":"41.27"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5-20","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0.5, 20 clients per round)","model":"FedASAM + SWA","rank_in_archive_order":1,"of":5,"metrics":{"ACC@1-100Clients":"48.27"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5-20","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0.5, 20 clients per round)","model":"FedASAM","rank_in_archive_order":2,"of":5,"metrics":{"ACC@1-100Clients":"47.78"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5-20","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0.5, 20 clients per round)","model":"FedSAM + SWA","rank_in_archive_order":3,"of":5,"metrics":{"ACC@1-100Clients":"46.47"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5-20","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0.5, 20 clients per round)","model":"FedSAM","rank_in_archive_order":4,"of":5,"metrics":{"ACC@1-100Clients":"46.05"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5-20","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0.5, 20 clients per round)","model":"FedAvg","rank_in_archive_order":5,"of":5,"metrics":{"ACC@1-100Clients":"42.17"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5-5","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0.5, 5 clients per round)","model":"FedASAM + SWA","rank_in_archive_order":1,"of":5,"metrics":{"ACC@1-100Clients":"49.17"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5-5","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0.5, 5 clients per round)","model":"FedSAM + SWA","rank_in_archive_order":2,"of":5,"metrics":{"ACC@1-100Clients":"47.96"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5-5","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0.5, 5 clients per round)","model":"FedASAM","rank_in_archive_order":3,"of":5,"metrics":{"ACC@1-100Clients":"45.61"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5-5","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0.5, 5 clients per round)","model":"FedSAM","rank_in_archive_order":4,"of":5,"metrics":{"ACC@1-100Clients":"44.73"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-0-5-5","task":"Federated Learning","dataset":"CIFAR-100 (alpha=0.5, 5 clients per round)","model":"FedAvg","rank_in_archive_order":5,"of":5,"metrics":{"ACC@1-100Clients":"40.43"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-1000-10","task":"Federated Learning","dataset":"CIFAR-100 (alpha=1000, 10 clients per round)","model":"FedASAM","rank_in_archive_order":1,"of":5,"metrics":{"ACC@1-100Clients":"54.97"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-1000-10","task":"Federated Learning","dataset":"CIFAR-100 (alpha=1000, 10 clients per round)","model":"FedASAM + SWA","rank_in_archive_order":2,"of":5,"metrics":{"ACC@1-100Clients":"54.79"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-1000-10","task":"Federated Learning","dataset":"CIFAR-100 (alpha=1000, 10 clients per round)","model":"FedSAM + SWA","rank_in_archive_order":3,"of":5,"metrics":{"ACC@1-100Clients":"53.67"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-1000-10","task":"Federated Learning","dataset":"CIFAR-100 (alpha=1000, 10 clients per round)","model":"FedSAM","rank_in_archive_order":4,"of":5,"metrics":{"ACC@1-100Clients":"53.39"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-1000-10","task":"Federated Learning","dataset":"CIFAR-100 (alpha=1000, 10 clients per round)","model":"FedAvg","rank_in_archive_order":5,"of":5,"metrics":{"ACC@1-100Clients":"50.25"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-1000-20","task":"Federated Learning","dataset":"CIFAR-100 (alpha=1000, 20 clients per round)","model":"FedASAM","rank_in_archive_order":1,"of":5,"metrics":{"ACC@1-100Clients":"54.5"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-1000-20","task":"Federated Learning","dataset":"CIFAR-100 (alpha=1000, 20 clients per round)","model":"FedSAM + SWA","rank_in_archive_order":2,"of":5,"metrics":{"ACC@1-100Clients":"54.36"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-1000-20","task":"Federated Learning","dataset":"CIFAR-100 (alpha=1000, 20 clients per round)","model":"FedASAM + SWA","rank_in_archive_order":3,"of":5,"metrics":{"ACC@1-100Clients":"54.1"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-1000-20","task":"Federated Learning","dataset":"CIFAR-100 (alpha=1000, 20 clients per round)","model":"FedSAM","rank_in_archive_order":4,"of":5,"metrics":{"ACC@1-100Clients":"53.97"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-1000-20","task":"Federated Learning","dataset":"CIFAR-100 (alpha=1000, 20 clients per round)","model":"FedAvg","rank_in_archive_order":5,"of":5,"metrics":{"ACC@1-100Clients":"50.66"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-1000-5","task":"Federated Learning","dataset":"CIFAR-100 (alpha=1000, 5 clients per round)","model":"FedASAM","rank_in_archive_order":1,"of":5,"metrics":{"ACC@1-100Clients":"54.81"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-1000-5","task":"Federated Learning","dataset":"CIFAR-100 (alpha=1000, 5 clients per round)","model":"FedSAM","rank_in_archive_order":2,"of":5,"metrics":{"ACC@1-100Clients":"54.01"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-1000-5","task":"Federated Learning","dataset":"CIFAR-100 (alpha=1000, 5 clients per round)","model":"FedSAM + SWA","rank_in_archive_order":3,"of":5,"metrics":{"ACC@1-100Clients":"53.9"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-1000-5","task":"Federated Learning","dataset":"CIFAR-100 (alpha=1000, 5 clients per round)","model":"FedASAM + SWA","rank_in_archive_order":4,"of":5,"metrics":{"ACC@1-100Clients":"53.86"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cifar-100-alpha-1000-5","task":"Federated Learning","dataset":"CIFAR-100 (alpha=1000, 5 clients per round)","model":"FedAvg","rank_in_archive_order":5,"of":5,"metrics":{"ACC@1-100Clients":"49.92"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cityscapes","task":"Federated Learning","dataset":"Cityscapes heterogeneous","model":"SiloBN + ASAM","rank_in_archive_order":1,"of":9,"metrics":{"mIoU":"49.75"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cityscapes","task":"Federated Learning","dataset":"Cityscapes heterogeneous","model":"SiloBN + SAM","rank_in_archive_order":2,"of":9,"metrics":{"mIoU":"49.1"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cityscapes","task":"Federated Learning","dataset":"Cityscapes heterogeneous","model":"SiloBN","rank_in_archive_order":3,"of":9,"metrics":{"mIoU":"45.96"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cityscapes","task":"Federated Learning","dataset":"Cityscapes heterogeneous","model":"FedSAM + SWA","rank_in_archive_order":4,"of":9,"metrics":{"mIoU":"43.42"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cityscapes","task":"Federated Learning","dataset":"Cityscapes heterogeneous","model":"FedASAM + SWA","rank_in_archive_order":5,"of":9,"metrics":{"mIoU":"43.02"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cityscapes","task":"Federated Learning","dataset":"Cityscapes heterogeneous","model":"FedAvg + SWA","rank_in_archive_order":6,"of":9,"metrics":{"mIoU":"42.48"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cityscapes","task":"Federated Learning","dataset":"Cityscapes heterogeneous","model":"FedASAM","rank_in_archive_order":7,"of":9,"metrics":{"mIoU":"42.27"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cityscapes","task":"Federated Learning","dataset":"Cityscapes heterogeneous","model":"FedSAM","rank_in_archive_order":8,"of":9,"metrics":{"mIoU":"41.22"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-cityscapes","task":"Federated Learning","dataset":"Cityscapes heterogeneous","model":"FedAvg","rank_in_archive_order":9,"of":9,"metrics":{"mIoU":"38.65"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-landmarks-user-160k","task":"Federated Learning","dataset":"Landmarks-User-160k","model":"FedASAM + SWA","rank_in_archive_order":1,"of":6,"metrics":{"Acc@1-1262Clients":"68.32"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-landmarks-user-160k","task":"Federated Learning","dataset":"Landmarks-User-160k","model":"FedSAM + SWA","rank_in_archive_order":2,"of":6,"metrics":{"Acc@1-1262Clients":"68.12"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-landmarks-user-160k","task":"Federated Learning","dataset":"Landmarks-User-160k","model":"FedAvg + SWA","rank_in_archive_order":3,"of":6,"metrics":{"Acc@1-1262Clients":"67.52"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-landmarks-user-160k","task":"Federated Learning","dataset":"Landmarks-User-160k","model":"FedASAM","rank_in_archive_order":4,"of":6,"metrics":{"Acc@1-1262Clients":"64.23"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-landmarks-user-160k","task":"Federated Learning","dataset":"Landmarks-User-160k","model":"FedSAM","rank_in_archive_order":5,"of":6,"metrics":{"Acc@1-1262Clients":"63.72"},"uses_additional_data":false},{"leaderboard":"/sota/federated-learning-on-landmarks-user-160k","task":"Federated Learning","dataset":"Landmarks-User-160k","model":"FedAvg","rank_in_archive_order":6,"of":6,"metrics":{"Acc@1-1262Clients":"61.91"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-cifar-100-alpha-0-20","task":"Image Classification","dataset":"CIFAR-100 (alpha=0, 20 clients per round)","model":"FedAvgM + ASAM + SWA","rank_in_archive_order":1,"of":1,"metrics":{"ACC@1-100Clients":"51.58"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2203.11834","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2203.11834"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/debcaldarola/fedsam","reach":{"status":"ok","spdx":"BSD-2-Clause"}}],"summary":{"ran":1},"by_repo_kind":{"official":{"samples":1,"ran":1,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"7690c4d309642d3f","entry":"get_metrics_names","repo":"debcaldarola/fedsam","repo_kind":"official","path":"models/metrics/writer.py","file_url":"https://github.com/debcaldarola/fedsam/blob/HEAD/models/metrics/writer.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-2-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"7690c4d309642d3f"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}