Skip to content

Latest commit

 

History

History
170 lines (113 loc) · 16.9 KB

tbd.md

File metadata and controls

170 lines (113 loc) · 16.9 KB
Raw

-Github: https://anonymous.4open.science/r/fedil

FederatedTrust: A Solution for Trustworthy Federated Learning

Thus, this work analyzes the existing requirements for trustworthiness evaluation in FL and proposes a comprehensive taxonomy of six pillars (privacy, robustness, fairness, explainability, accountability, and federation) with notions and more than 30 metrics for computing the trustworthiness of FL models. Then, an algorithm called FederatedTrust has been designed according to the pillars and metrics identified in the previous taxonomy to compute the trustworthiness score of FL models. A prototype of FederatedTrust has been implemented and deployed into the learning process of FederatedScope, a well-known FL framework.

Welfare and Fairness Dynamics in Federated Learning

To address this problem, we designed a novel incentive mechanism that involves a client selection process to remove low-quality clients and a money transfer process to ensure a fair reward distribution. Our experimental results strongly demonstrate that the proposed incentive mechanism can effectively improve the duration and fairness of the federation.

Must-read

xyz

xyz

FedRL

  • e λ is the testing accuracy of the global models on the validation set after t rounds

BC

FL Client Selection

Asynch FL

FL

rPPG

FedFOR

FedFOR: Stateless Heterogeneous Federated Learning with First-Order Regularization, https://arxiv.org/abs/2209.10537

We derive a first-order gradient regularization to penalize inconsistent local updates due to local data heterogeneity. Specifically, to mitigate weight divergence, we introduce a first-order approximation of the global data distribution into local objectives, which intuitively penalizes updates in the opposite direction of the global update. The end result is a stateless FL algorithm that achieves 1) significantly faster convergence (i.e., fewer communication rounds) and 2) higher overall converged performance than SOTA methods under non-IID data distribution. Importantly, our approach does not impose unrealistic limits on the client size, enabling learning from a large number of clients as is typical in most FL applications.

Our code will be released at https://github.com/GT-RIPL/FedFOR.

Most FL algorithms can be categorized under innovation to one of the components.

  • ServerOpt methods improve convergence by adding server-side momentum or adaptive optimizers. FedAvgM (Hsu, Qi, and Brown 2019) and SlowMo (Wang et al. 2019) simulates imbalanced data distribution, which is one form of data heterogeneity, and proposes to use server side momentum to improve convergence. FedAdagrad/FedYogi/FedAdam (Reddi et al. 2020) extends FedAvg by including several adaptive optimizers on the server-side to combat data heterogeneity.
  • ClientOpt methods add client-side regularization to reduce the effects of data heterogeneity. FedProx (Li et al. 2020) proposes to add a proximal term (L2 regularization) to limit the impact of non-IID data. FedCurv (Shoham et al. 2019) adapts a second-order gradient regularization method, EWC (Kirkpatrick et al. 2017), from continual learning to FL. Recent works, FedPD (Zhang et al. 2020) and FedDyn (Acar et al. 2021) improve convergence on non-IID data by including a first-order regularization term, which seeks consensus among clients. ClientOpt is investigated more heavily than ServerOpt because poor performance due to data heterogeneity is a direct result of local optimization. Furthermore, ClientOpt is more difficult to design because of the stateless requirement for FL when the number of clients is large. For example, FedPD and FedDyn are stateful algorithms and can be difficult to apply to large-scale, real-world FL settings.

In this paper, we focus on the ClientOpt component to improve convergence under non-IID data distribution and compare analytically and empirically to all ClientOpt methods mentioned in this section.