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

Fair and Efficient Contribution Valuation for Vertical Federated Learning

ICLR 2024 · Federated learning and privacy › Personalized and heterogeneous FL · approach: Vertical federated learning · 14 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 VerFedSV, a retraining-free Shapley-based contribution valuation for vertical federated learning that uses embeddings across training time stamps, with low-rank matrix completion for synchronous VFL and direct computation for asynchronous VFL.

Problem

Fairly valuing each client's contribution in vertical federated learning, where the Shapley value requires prohibitively many retrained models and the global model is a concatenation of unshared local models.

Approach

Define a model-dependent per-time-stamp utility from clients' embeddings (latest embeddings for participating clients, previous ones otherwise), average Shapley values over time stamps (equivalent to classic Shapley on averaged utility), fill missing embeddings in synchronous VFL via low-rank matrix completion, and in asynchronous VFL reflect each client's computational/communication strength.

Key results

• Relative VerFedSV difference is near 0 for identical-feature clients (exactly 0 on Adult, near 0 on Web, Covtype, RCV1) and grows with feature heterogeneity.
• Randomly-featured artificial clients receive much lower valuations, e.g. under 2.4% each of total VerFedSV in the synchronous setting on all datasets.
• Kendall's rank correlation between SHAP and VerFedSV is above 0.6 on all four datasets (e.g. 1.0 Adult, 0.94 Covtype synchronous).

Measured effects reported in the paper

Surprising findings

• Adding clients with identical features increases the total VerFedSV, which the authors note could let clients cheat by duplicating data.

Limitations

• Adding clients with identical features inflates total valuation, enabling potential cheating.
• Exact computation is exponential in the number of clients; experiments rely on Monte-Carlo sampling for M > 10.
• Experiments use only linear or small MLP models on four LIBSVM datasets.

Experimental setup

Datasets: Adult Web Covtype RCV1

Benchmarks: Adult Web Covtype RCV1

Metrics: percentage of VerFedSV relative VerFedSV difference Kendall's rank correlation run time rank of embedding matrix

Baselines: SHAP

Builds on: Shapley value federated Shapley value (FedSV) FedSGD vertical asynchronous federated learning (VAFL) low-rank matrix completion

Closest papers by meaning

• You Get What You Give: Reciprocally Fair Federated Learning (ICML 2025 · similarity 0.86)
• TurboSVM-FL: Boosting Federated Learning through SVM Aggregation for Lazy Clients (AAAI 2024 · similarity 0.86)
• Data Valuation and Detections in Federated Learning (CVPR 2024 · similarity 0.86)
• FairTrade: Achieving Pareto-Optimal Trade-Offs between Balanced Accuracy and Fairness in Federated Learning (AAAI 2024 · similarity 0.85)
• DU-Shapley: A Shapley Value Proxy for Efficient Dataset Valuation (NeurIPS 2024 · similarity 0.85)
• Fair Federated Learning under Domain Skew with Local Consistency and Domain Diversity (CVPR 2024 · similarity 0.85)