Abstract
Federated learning (FL), as an effective decentralized distributed learning approach, enables multiple institutions to jointly train a model without sharing their local data. However, the domain feature shift caused by different acquisition devices/clients substantially degrades the performance of the FL model. Furthermore, most existing FL approaches aim to improve accuracy without considering reliability (e.g., confidence or uncertainty). The predictions are thus unreliable when deployed in safety-critical applications. Therefore, aiming at improving the performance of FL in non-Domain feature issues while enabling the model more reliable. In this paper, we propose a novel trusted federated disentangling network, termed TrFedDis, which utilizes feature disentangling to enable the ability to capture the global domain-invariant cross-client representation and preserve local client-specific feature learning. Meanwhile, to effectively integrate the decoupled features, an uncertainty-aware decision fusion is also introduced to guide the network for dynamically integrating the decoupled features at the evidence level, while producing a reliable prediction with an estimated uncertainty. To the best of our knowledge, our proposed TrFedDis is the first work to develop an FL approach based on evidential uncertainty combined with feature disentangling, which enhances the performance and reliability of FL in non-IID domain features. Extensive experimental results show that our proposed TrFedDis provides outstanding performance with a high degree of reliability as compared to other state-of-the-art FL approaches.
Abstract (translated)
联邦学习(FL),作为一种有效的分布式分散学习方法,使多个机构共同训练模型而无需共享本地数据。然而,由不同 acquisition 设备/客户导致 domain 特征 shift 显著削弱了 FL 模型的性能。此外,大多数现有的 FL 方法旨在提高准确性而不考虑可靠性(例如,信心或不确定性)。因此在安全关键应用中部署的预测的可靠性较低。因此,旨在改善 FL 在非 domain 特征问题上的表现,同时使模型更加可靠。在本文中,我们提出了一种新的可信的联邦解离网络,名为 TrFedDis,它利用特征解离能力,实现能够捕捉全球 domain 不变的跨客户代表并保留本地客户特定特征学习的能力。同时,为了有效地整合解离的特征,引入一种不确定性意识的决策融合,以指导网络在证据级别上动态整合解离的特征,同时产生估计的不确定性可靠的预测。据我们所知,我们提出的 TrFedDis 是首先基于证据不确定性和特征解离开发的一种 FL 方法,从而提高了 FL 在非 IID domain 特征问题上的表现和可靠性。广泛的实验结果表明,我们提出的 TrFedDis 提供了出色的表现,与 other 先进的 FL 方法相比,具有极高的可靠性。
URL
https://arxiv.org/abs/2301.12798