Skip to main content
QUICK REVIEW

[论文解读] Cumulative Utility Parity for Fair Federated Learning under Intermittent Client Participation

Stefan Kambiz Behfar, Richard Mortier|arXiv (Cornell University)|Feb 14, 2026
Privacy-Preserving Technologies in Data被引用 0
一句话总结

论文在联邦学习中定义并优化可用性归一化的累计效用,以在客户端参与间歇的情况下实现长期公平性,采用逆可用性采样和代理更新,在基线方法上取得强劲的经验提升。

ABSTRACT

In real-world federated learning (FL) systems, client participation is intermittent, heterogeneous, and often correlated with data characteristics or resource constraints. Existing fairness approaches in FL primarily focus on equalizing loss or accuracy conditional on participation, implicitly assuming that clients have comparable opportunities to contribute over time. However, when participation itself is uneven, these objectives can lead to systematic under-representation of intermittently available clients, even if per-round performance appears fair. We propose cumulative utility parity, a fairness principle that evaluates whether clients receive comparable long-term benefit per participation opportunity, rather than per training round. To operationalize this notion, we introduce availability-normalized cumulative utility, which disentangles unavoidable physical constraints from avoidable algorithmic bias arising from scheduling and aggregation. Experiments on temporally skewed, non-IID federated benchmarks demonstrate that our approach substantially improves long-term representation parity, while maintaining near-perfect performance.

研究动机与目标

  • 在现实世界的、客户端参与模式不稳定的情况下,激发对FL的公平性关注。
  • 将累计效用对参与机会随时间的影响进行定义和形式化,作为公平性标准。
  • 开发机制(可用性建模、逆可用性采样和代理更新),在不牺牲性能的前提下实现长期公平性。
  • 给出理论保证,显示在间歇参与下累计效用差距的收敛性质。
  • 在非IID基准和随时间偏斜的参与模式下,实证验证该方法相对于基线的性能与公平性。

提出的方法

  • 引入时间维度的效用跟踪,以在训练轮次上维护每个客户端的累计效用。
  • 通过长期可用性对效用进行归一化,以计算可用性归一化效用。
  • 实现具备逆可用性权重的自适应采样,以实现长期参与机会的均等化。
  • 纳入表示感知的缺失客户端代理更新,以缓解数据缺失带来的影响。
  • 给出理论结果(引理1、定理1、引理2–3、定理3),关于公平性和代理影响的证明。
  • 在具有现实可用性轨迹的非IID CIFAR-10基准上进行评估,比较对象为q-FFL和PHP-FL。
Figure 1 : Distribution of device availability percentages for the trace data (Yang et al. , 2021 ) , where device availability percentage is defined as the percentage of time between the first and last times a device was seen to be live and available to perform FL, i.e., was charging and connected
Figure 1 : Distribution of device availability percentages for the trace data (Yang et al. , 2021 ) , where device availability percentage is defined as the percentage of time between the first and last times a device was seen to be live and available to perform FL, i.e., was charging and connected

实验结果

研究问题

  • RQ1在客户端参与不稳定且与数据/资源约束相关时,如何定义FL中的公平性?
  • RQ2可用性归一化的累计效用是否收敛,是否可以通过逆可用性采样有效地实现平衡?
  • RQ3对于不可用客户端的代理更新是否能在 dropout 模式下保持模型质量与公平性?
  • RQ4与现有公平性方法(如q-FFL、PHP-FL)相比,该方法在长期表示平等和性能方面的表现如何?

主要发现

MethodAvg AccJain (Acc)Utility CVJain (Utility)Sel. GapGini
q-FFL60.10.720.640.420.800.35
PHP-FL67.710.800.420.780.520.20
Ours (no surrogate)80.430.9750.280.880.310.04
Ours (with surrogate)80.430.9750.190.940.310.04
  • 可用性归一化的累计效用减少了客户端之间的长期差异,达到表示平等。
  • 逆可用性采样使长期客户端参与频率趋于均等,即使在可用性异质的情况下也如此。
  • 缺失客户端的代理更新缓解了表示丢失并在 dropout 下维持学习连续性。
  • 实证结果显示近乎完美的准确性公平性(Jain_acc = 0.975)和显著低的效用变异系数(Utility CV = 0.19),相比基线表现更优。
  • 与q-FFL和PHP-FL相比,所提方法获得更高的平均准确性(Avg Acc = 80.43)和更好的公平性指标(选择差距和Gini更低)。
  • 在引入代理更新的版本中,平均准确性和Jain_acc与无代理版本保持相同,同时提升了效用CV和效用的Jain指标。
Figure 2 : Accuracy and fairness variance versus round number using random sampling.
Figure 2 : Accuracy and fairness variance versus round number using random sampling.

更好的研究,从现在开始

从论文设计到论文写作,大幅缩短您的研究时间。

无需绑定信用卡

本解读由 AI 生成,并经人工编辑审核。