FedMark-FM: Auditable, Risk-Adjusted Data Markets for Federated Foundation-Model Adaptation

2026-07-08

AI Summary

This work addresses the limitations of existing federated learning incentive mechanisms, which assume client homogeneity and fail to fairly evaluate heterogeneous, non-IID, workflow-dependent, and privacy-constrained contributions of diverse private artifacts in foundation model fine-tuning, rendering them vulnerable to strategic attacks. To overcome this, we propose FedMark-FM, the first auditable incentive framework for federated foundation models, modeling clients as sellers of multi-type artifacts. We introduce S3Val, a hierarchical uncertainty-aware Shapley estimator that supports sequential workflow valuation, and integrate it with a confidence lower bound to construct a risk-adjusted, budget-feasible payment mechanism. We theoretically prove that sequential workflow valuation is the unique credit assignment rule satisfying service causality. Experiments on FEVER retrieval, RAG generation, and LoRA fine-tuning tasks demonstrate that our approach improves downstream accuracy by 7.5–8.1 points under prompt injection attacks, effectively filters out strategic clients, protects rare experts, and achieves extremely narrow split conformal prediction intervals (average width 0.0141).

Abstract

Federated foundation-model adaptation increasingly relies on heterogeneous private artifacts (retrieval corpora, prompts and demonstrations, LoRA adapters, preference and safety data, and update sketches), yet existing federated-learning incentive mechanisms price clients as homogeneous data or update providers. This assumption poorly matches foundation-model pipelines, where contribution value is heterogeneous, non-IID, pipeline-dependent, privacy-constrained, and vulnerable to strategic behavior. We propose FedMark-FM, an auditable, risk-adjusted data-market framework that models clients as sellers of typed artifacts, estimates marginal contribution with S3Val, a stratified, uncertainty-aware Shapley estimator supporting pipeline-ordered valuation, and converts lower-confidence-bound values into budget-feasible payments penalizing duplication, sybil splitting, poisoned adapters, privacy-budget gaming, and cost inflation. We evaluate FedMark-FM-Bench across FEVER retrieval, held-out generator-backed RAG, and trained PEFT/LoRA tracks. Under a held-out prompt-injection poisoner, FedMark-FM improves downstream accuracy by 7.5-8.1 points over volume, leave-one-out, and FL-Shapley while selecting zero strategic clients. Split-conformal calibration reaches full lower-bound coverage at mean width 0.0141, versus 0.33 for naive intervals. We prove pipeline-ordered valuation is the unique credit rule respecting serving causality, and show it materially changes credit assignment (Spearman 0.76, selected-set overlap 0.67) while leaving held-out task quality unchanged; the market preserves rare specialists with audit-ready ledgers at 200-1000-client scale. FedMark-FM shows incentives for federated foundation models can be engineered as auditable data infrastructure coupling valuation, mechanism design, privacy interfaces, and pipeline-order semantics.

Problem

• federated learning
• foundation models
• incentive mechanisms
• heterogeneous contributions
• data markets

Innovation

• Federated Foundation Models
• Pipeline-Ordered Valuation
• Risk-Adjusted Data Markets
• S3Val
• Auditable Incentives