Abstract:Federated learning (FL) has emerged as a pivotal approach in machine learning, enabling multiple participants to collaboratively train a global model without sharing raw data. While FL finds applications in various domains such as healthcare and finance, it is challenging to ensure global model convergence when participants do not contribute equally and/or honestly. To overcome this challenge, principled mechanisms are required to evaluate the contributions made by individual participants in the FL setting. Existing solutions for contribution assessment rely on general accuracy evaluation, often failing to capture nuanced dynamics and class-specific influences. This paper proposes a novel contribution assessment method called ShapFed for fine-grained evaluation of participant contributions in FL. Our approach uses Shapley values from cooperative game theory to provide a granular understanding of class-specific influences. Based on ShapFed, we introduce a weighted aggregation method called ShapFed-WA, which outperforms conventional federated averaging, especially in class-imbalanced scenarios. Personalizing participant updates based on their contributions further enhances collaborative fairness by delivering differentiated models commensurate with the participant contributions. Experiments on CIFAR-10, Chest X-Ray, and Fed-ISIC2019 datasets demonstrate the effectiveness of our approach in improving utility, efficiency, and fairness in FL systems. The code can be found at <a class="link-external link-https" href="https://github.com/tnurbek/shapfed" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

The paper primarily aims to address several key issues in Federated Learning (FL): 1. **Inequality and Dishonest Contributions**: In federated learning, when participants (clients) contribute unequally or dishonestly to the global model, it can lead to difficulties in the convergence of the global model. Therefore, a method is needed to evaluate the actual contribution of each participant. 2. **Limitations of Existing Contribution Evaluation Methods**: Current methods mostly rely on accuracy evaluation on an auxiliary validation set to measure contributions. This approach often fails to capture the nuances of participants' impacts, especially in cases of imbalanced data or significant class differences. ### Solution The paper proposes a new contribution evaluation method—ShapFed, and a weighted aggregation strategy based on this method—ShapFed-WA, aimed at addressing the aforementioned issues: - **ShapFed**: This method uses Shapley values from cooperative game theory to quantify each participant's contribution to the global model, particularly their influence on specific classes. Specifically, ShapFed focuses on the last layer parameters (gradients) to provide computational efficiency and more accurately assess the true Shapley values. - **ShapFed-WA**: This is a weighted aggregation method that aggregates model updates from different participants based on the contribution evaluation values obtained from ShapFed, thereby improving model performance in scenarios with imbalanced data. Additionally, to enhance the collaborative fairness of the system, the paper proposes a personalized update strategy where the server sends personalized model updates to each participant based on their contribution, ensuring that high-contributing participants receive better model performance. ### Experimental Results The paper conducts experiments on the CIFAR-10, Chest X-Ray, and Fed-ISIC2019 datasets to demonstrate the effectiveness of ShapFed and its weighted aggregation method. The experimental results show that these methods not only improve the utility and efficiency of the model but also enhance the fairness of the federated learning system.

Redefining Contributions: Shapley-Driven Federated Learning

Evaluate the Contribution of Multiple Participants in Federated Learning

FairFed: Improving Fairness and Efficiency of Contribution Evaluation in Federated Learning via Cooperative Shapley Value

A Fair Contribution Measurement Method for Federated Learning

Optimizing Federated Learning on Non-IID Data Using Local Shapley Value.

Efficient Participant Contribution Evaluation for Horizontal and Vertical Federated Learning

Don't Forget What I did?: Assessing Client Contributions in Federated Learning

DPVS-Shapley:Faster and Universal Contribution Evaluation Component in Federated Learning

ShapleyFL: Robust Federated Learning Based on Shapley Value

FedLaw: Value-Aware Federated Learning with Individual Fairness and Coalition Stability

Shapley-value-based Contribution Evaluation in Federated Learning: A Survey

Mitigating federated learning contribution allocation instability through randomized aggregation

Contributions Estimation in Federated Learning: A Comprehensive Experimental Evaluation

GTG-Shapley: Efficient and Accurate Participant Contribution Evaluation in Federated Learning

Towards Fair, Robust and Efficient Client Contribution Evaluation in Federated Learning

Improving Fairness for Data Valuation in Horizontal Federated Learning

Fast, Robust and Interpretable Participant Contribution Estimation for Federated Learning

Contribution Evaluation in Federated Learning: Examining Current Approaches

Contribution Evaluation of Heterogeneous Participants in Federated Learning via Prototypical Representations

A Principled Approach to Data Valuation for Federated Learning

Fair and efficient contribution valuation for vertical federated learning