Research map / Federated learning and privacy / Communication-efficient FL
FedCross: Intertemporal Federated Learning Under Evolutionary Games
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 FedCross, a two-stage incentive framework for federated learning under user mobility that combines a multi-objective (genetic-style) task migration algorithm, evolutionary game dynamics for region formation, and a greedy procurement auction with critical-value payments.
Problem
Federated learning with mobile users who frequently leave base-station coverage, causing interrupted training tasks, high migration/communication overhead, and unsustainable participation incentives under limited budgets.
Approach
Combines task migration via an NSGA-style multi-objective evolutionary algorithm (SBX crossover, polynomial mutation) with evolutionary game theory (replicator dynamics) to predict user region choices, and a greedy procurement auction with critical-value payments to reward base stations that provide high-quality models.
Key results
- On MNIST and CIFAR-10 with geospatial augmentation, FedCross is reported to achieve the highest accuracy among compared methods (FedCross > WCNFL > SAVFL > BasicFL), per qualitative discussion without precise numbers in the text.
- Auction-based payment algorithm is reported to yield a more stable and lower payment cost than the non-payment variant over time.
- Evolutionary game simulations show population proportions across three areas converging to dynamic stability (e.g., [0.25, 0.35, 0.4]) after around time 300.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use FedCross instead of BasicFL, SAVFL, or WCNFL | FedCross · MNIST | accuracy | Helped |
| Use FedCross instead of BasicFL, SAVFL, or WCNFL | FedCross · CIFAR-10 | accuracy | Helped |
| Use the proposed multi-objective migration algorithm instead of random search or simulated annealing | FedCross migration algorithm · Task allocation | task allocation performance | Helped |
| Use FedCross's auction allocation rule instead of the BasicFL traditional auction rule or WCNFL reverse auction | FedCross auction · Auction phase | payment cost | Helped |
| Offer tangible rewards through base-station auctions | FedCross auction · User participation in training | participation | Helped |
| Add the threshold-based payment algorithm rather than omit payment (ablation) | FedCross payment algorithm · Auction phase | payment cost stability | Helped |
Limitations
- Experimental evaluation is limited to simple image datasets (MNIST, CIFAR-10) and simulated mobility, with no precise numerical results reported in the text.
- Stability/equilibrium theorems rely on loosely justified arguments (e.g., Lyapunov derivative argument and boundedness lemma) and are not rigorously proven.
- Auction payment and incentive-compatibility claims are asserted with informal proofs.
Experimental setup
Datasets: MNIST CIFAR-10
Benchmarks: MNIST CIFAR-10
Metrics: model accuracy payment cost communication overhead task allocation performance
Baselines: BasicFL SAVFL WCNFL
Builds on: hierarchical federated learning replicator dynamics greedy procurement auction NSGA-II-style multi-objective evolutionary algorithm
Closest papers by meaning
- Federated Behavioural Planes: Explaining the Evolution of Client Behaviour in Federated Learning (NeurIPS 2024 · similarity 0.86)
- FedCompetitors: Harmonious Collaboration in Federated Learning with Competing Participants (AAAI 2024 · similarity 0.86)
- Communication-Efficient Federated Non-Linear Bandit Optimization (ICLR 2024 · similarity 0.85)
- You Get What You Give: Reciprocally Fair Federated Learning (ICML 2025 · similarity 0.85)
- Multiplayer Federated Learning: Reaching Equilibrium with Less Communication (NeurIPS 2025 · similarity 0.85)
- Efficient Adaptive Federated Optimization (NeurIPS 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.