联邦学习贡献评估综述

数据不动的联邦学习框架是多个数据持有方合作训练机器学习模型的新范式.多个数据持有方参与联邦学习时的贡献评估是联邦学习的核心问题之一.参与方贡献评估需要兼顾有效性、公平性和合理性等要素,在理论方法与实际应用中均面临多项挑战.贡献评估首先需要明确如何度量数据价值,然而数据估值存在主观性与依赖于实际任务场景的特点,如何设计有效、可靠并对恶意数据鲁棒的数据估值指标是第一大挑战.其次,联邦学习合作中的参与方贡献评估是经典的合作博弈问题,如何制定公平合理的参与方贡献评估方案,实现参与方一致认可的博弈平衡是第二大挑战.最后,参与方贡献评估往往计算复杂度高,同时,联邦学习中围绕模型的数据估值时间开销大,因此,在实践中如何设计高效且准确的近似算法是第三大挑战.近年来,为了有效地解决上述挑战,学术界对联邦学习中的贡献评估问题展开了广泛的研究.首先,简要介绍联邦学习与参与方贡献评估的背景知识;然后,综述数据估值指标、参与方贡献评估方案和相关优化技术;最后,讨论了联邦学习贡献评估仍面临的挑战并展望未来研究的发展方向.

Survey on Contribution Evaluation for Federated Learning

Federated learning is a collaborative machine learning framework with multiple participants whose training datasets are kept locally.How to evaluate the corresponding data contribution of each participant is one of the critical problems of federated learning.However,contribution evaluation in federated learning faces multiple challenges.First,to evaluate participant contribution,data value needs to be quantified,however,data valuation is challenging because it is subjective,task context-dependent and vulnerable to malicious data.Second,participant contribution evaluation is a classic cooperative game problem,and a fair yet rational cooperative contribution evaluation scheme is needed to achieve an optimal equilibrium among all participants.In recent years,researchers have conducted extensive studies on participant contribution evaluation in federated learning to tackle the above challenges.This paper first introduces the background knowledge of federated learning and contribution evaluation.Then,data valuation metrics,contribution evaluation schemes,and corresponding optimization technologies are surveyed successively.Finally,the remaining challenges of contribution evaluation and potential future work are discussed.