Research map / Federated learning and privacy / Privacy attacks and secure learning

Reputation-aware Revenue Allocation for Auction-based Federated Learning

AAAI 2025 · Federated learning and privacy › Privacy attacks and secure learning · approach: Private federated aggregation · 4 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 ARAS-AFL, a Lyapunov-optimization-based dynamic revenue allocation strategy that lets an auction-based federated learning marketplace manage its reputation and attract data owners in a competitive multi-marketplace ecosystem.

Problem

Auctioneer decision support in auction-based federated learning assumes a single monopolistic marketplace; in reality multiple AFL marketplaces compete for the same data owners, so auctioneers need strategies to attract DOs while protecting profit.

Approach

Frames DO attraction as reputation management: a virtual regret queue tracks payouts that fail to increase DO count, and Lyapunov drift minimization yields a closed-form dynamic revenue-sharing ratio balancing profit and regret, using Beta-reputation-system reputation scores.

Key results

• Under the overrelease scenario, ARAS-AFL improves over the best baseline by 64.80% in total revenue, 150.71% in number of DOs, 13.63% in public reputation, and 4.34% in FL accuracy (Table 1 averages).
• Under the underrelease scenario, ARAS-AFL improves by 33.32%, 46.67%, 69.98%, and 5.20% in revenue, DO count, public reputation, and accuracy respectively.
• The abstract reports outperforming the best baseline by 49.06%, 98.69%, 10.32%, and 4.77% across the four metrics.

Measured effects reported in the paper

Surprising findings

• Public reputation of ARAS-AFL shows a decreasing trend over the experiments, as attracting more DOs leaves fewer high-payoff tasks for them.
• Random allocation shows less reputation fluctuation than ARAS-AFL in the experiments.

Limitations

• Evaluation relies on simulated marketplaces and synthetic task payments rather than real-time FL task data.
• Only two simple baselines (random and greedy) are compared, as no prior auctioneer-focused decision support method exists.
• Reputation scores decline over time for ARAS-AFL, a trade-off the paper notes but does not fully resolve.

Experimental setup

Datasets: MNIST CIFAR-10 Fashion-MNIST EMNIST-digits EMNIST-letters Kuzushiji-MNIST

Benchmarks: MNIST CIFAR-10 Fashion-MNIST EMNIST-digits EMNIST-letters Kuzushiji-MNIST

Metrics: averaged total revenue averaged number of DOs public reputation FL model accuracy

Baselines: Random revenue allocation (Rand) Greedy revenue allocation (Greedy)

Builds on: Lyapunov optimization Beta Reputation System FedAvg PAS-AFL

Closest papers by meaning

• Multi-Session Budget Optimization for Forward Auction-based Federated Learning (ICML 2025 · similarity 0.91)
• DualGFL: Federated Learning with a Dual-Level Coalition-Auction Game (AAAI 2025 · similarity 0.87)
• Auctionformer: A Unified Deep Learning Algorithm for Solving Equilibrium Strategies in Auction Games (ICML 2024 · similarity 0.86)
• AuctionNet: A Novel Benchmark for Decision-Making in Large-Scale Games (NeurIPS 2024 · similarity 0.86)
• Learning Bayesian Nash Equilibrium in Auction Games via Approximate Best Response (ICML 2025 · similarity 0.85)
• Increasing Revenue in Efficient Combinatorial Auctions by Learning to Generate Artificial Competition (AAAI 2025 · similarity 0.85)