Research map / Federated learning and privacy / Personalized and heterogeneous FL

Fair Federated Survival Analysis

AAAI 2025 · Federated learning and privacy › Personalized and heterogeneous FL · approach: Federated learning with pretrained models · 2 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes FairFSA, a federated survival analysis framework that combines federated pseudo values, distributionally robust optimization (PseudoDRO) for local fairness, and concordance-index-weighted FedAvg aggregation for global fairness.

Problem

Federated survival analysis models can produce biased predictions across demographic groups and clients because of demographic and censoring disparities, and existing fair survival methods require centralized pairwise comparisons or sensitive attributes.

Approach

Reformulates survival as regression on federated pseudo values so the loss is separable, enabling DRO to penalize worst-case subpopulation errors without sensitive attributes, and aggregates client models weighted by validation C-index to promote global fairness.

Key results

• On SUPPORT centralized, PseudoDRO reaches Favg 0.000 vs. 0.002-0.075 for baselines while C-Index is 0.61 (tied for best).
• In federated SUPPORT, FairFSA achieves C-Index 0.61 and Favg 0.001, the lowest fairness gap among compared FSA methods.
• On MSK-MET global fairness analysis, CI-weighted aggregation lowers the cross-client std of C-Index from 0.04 (FedAvg) to 0.03 and FI std from 0.005 to 0.003.

Measured effects reported in the paper

Surprising findings

• Pseudo-value-based models are reported to be far less affected by censoring than non-pseudo-value models, which vary widely across clients.

Limitations

• Covariate-dependent censoring and its dependence on sensitive attributes is out of scope.
• IBS (calibration) is not improved and is a stated weakness.
• Fairness-accuracy trade-off parameter alpha is selected by intersection heuristic rather than principled tuning.

Experimental setup

Datasets: FLChain SUPPORT SEER MSK-MET

Benchmarks: FLChain SUPPORT SEER MSK-MET

Metrics: C-Index IBS MAE-Uncensored MAE-Hinge individual fairness group fairness intersectional fairness Favg standard deviation across clients

Baselines: CoxPH DeepSurv DeepHit DeepPseudo FairSurv FIDP DRO-Cox FedCox FedDP Fed-DeepHit FedPseudo Fair-FedCox

Builds on: FedAvg Federated Pseudo Values Distributionally Robust Optimization DRO-Cox FIDP

Closest papers by meaning

• FedFACT: A Provable Framework for Controllable Group-Fairness Calibration in Federated Learning (NeurIPS 2025 · similarity 0.86)
• LoGoFair: Post-Processing for Local and Global Fairness in Federated Learning (AAAI 2025 · similarity 0.86)
• FairTrade: Achieving Pareto-Optimal Trade-Offs between Balanced Accuracy and Fairness in Federated Learning (AAAI 2024 · similarity 0.85)
• Multi-Dimensional Fair Federated Learning (AAAI 2024 · similarity 0.85)
• FedAA: A Reinforcement Learning Perspective on Adaptive Aggregation for Fair and Robust Federated Learning (AAAI 2025 · similarity 0.85)
• Does Egalitarian Fairness Lead to Instability? The Fairness Bounds in Stable Federated Learning Under Altruistic Behaviors (NeurIPS 2024 · similarity 0.85)