{"paper":{"title":"Federated Distillation for Whole Slide Image via Gaussian-Mixture Feature Alignment and Curriculum Integration","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Local Gaussian-mixture alignment of synthetic features lets pathology institutions train models together without exchanging parameters.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Cong Cong, Luru Jing, Yanyuan Chen, Yongzhi Cao","submitted_at":"2026-05-01T11:25:54Z","abstract_excerpt":"Federated learning (FL) offers a promising framework for collaborative digital pathology by enabling model training across institutions. However, real-world deployments face heterogeneity arising from diverse multiple instance learning (MIL) architectures and heterogeneous feature extractors across institutions. We propose FedHD, a novel FL framework that performs local Gaussian-mixture feature alignment tailored for WSI analysis. Instead of exchanging model parameters, each client independently distills semantically rich synthetic feature representations aligned with the distribution of real "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments on TCGA-IDH, CAMELYON16, and CAMELYON17 show that FedHD consistently outperforms state-of-the-art federated and distillation baselines.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The one-to-one distillation strategy and curriculum integration preserve diagnostic diversity and produce net positive transfer without introducing distribution shift that harms local performance.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"FedHD performs federated distillation for whole slide images by generating one synthetic feature set per real slide via Gaussian-mixture alignment and adding them via curriculum integration, outperforming prior federated and distillation methods on TCGA-IDH, CAMELYON16, and CAMELYON17.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Local Gaussian-mixture alignment of synthetic features lets pathology institutions train models together without exchanging parameters.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"7d12ccc72af0452823b53bfeb84913b982e48986e98d52a9f9f12739ac403f6d"},"source":{"id":"2605.00578","kind":"arxiv","version":2},"verdict":{"id":"70ef8e35-6400-4973-a9cd-048330c08e2b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-09T19:10:50.765147Z","strongest_claim":"Experiments on TCGA-IDH, CAMELYON16, and CAMELYON17 show that FedHD consistently outperforms state-of-the-art federated and distillation baselines.","one_line_summary":"FedHD performs federated distillation for whole slide images by generating one synthetic feature set per real slide via Gaussian-mixture alignment and adding them via curriculum integration, outperforming prior federated and distillation methods on TCGA-IDH, CAMELYON16, and CAMELYON17.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The one-to-one distillation strategy and curriculum integration preserve diagnostic diversity and produce net positive transfer without introducing distribution shift that harms local performance.","pith_extraction_headline":"Local Gaussian-mixture alignment of synthetic features lets pathology institutions train models together without exchanging parameters."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.00578/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_compliance","ran_at":"2026-05-19T18:02:37.750277Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"82e1ff267182ce1bb5916dfcfc63007ab9ec1dfd058ef20565a049052ab8c0d5"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}