一、摘要这篇文章提出的FedMGDA+有两方面的贡献,一方面能够抵御某些device的恶意攻击,具有鲁棒性;另一方面,在multi-objective optimization的框架下,不牺牲任何一个参与者的利益,达到帕累托最优。 二、问题…
Different from model quantization and partial model uploads presented in the previous chapter, evolutionary federated learning, more specifically, evolutionary federated neural architecture search, aims to optimize the architecture of neural network models, thereby reducing the communication costs caused by ...
(1)横向联邦学习。横向联邦学习又称为特征对齐的联邦学习(Feature-Aligned Federated Learning),是把...
Differential Privacy (DP) was not considered in experiment series 1 since the objective was to study the effects of data size, distribution, and the number of clients on the performance of distributed learning/federated learning in general. In experiment series 2, we compared the performance of ...
Federated Optimization: Distributed Machine Learning for On-Device Intelligence Stochastic, Distributed and Federated Optimization for Machine Learning. FL PhD Thesis. By Jakub Collaborative Deep Learning in Fixed Topology Networks Federated Multi-Task Learning Local Stochastic Approximation: A Unified View of...
Dragi Kimovski, Nishant Saurabh, Sandi Gec, Vlado Stankovski, and Radu Prodan. Multi-objective optimization framework for vmi distribu- tion in federated cloud repositories. In European Conference on Parallel Processing, pages 236-247. Springer, 2016....
In this section, we are going to consider Continual Learning (CL) problems, which involve the training of models over time. In the standard ML setting, the objective is to build a prediction model using a certain amount of data. A key point to discuss is that the training dataset is typi...
Here, ηt is called the learning rate. This update rule requires a single example (xi,yi), and in each update we can choose a random example. In this study we use logistic regression as our machine learning model, where the specific form of the objective function is given by (3)J(w...
《2022-IEEE-FedMGDA+: Federated Learning meets Multi-objective Optimization》同样关注性能最差的设备,为了兼顾公平性和鲁棒性,提出了 FedMGDA+方法。通过修改参与者梯度合并的权重来改进联邦模型公平性。多目标优化更关心当前模型在所有参与者的结果,它使用帕累托稳定(Pareto-stationary)解决方案。《2021-NIPS-...
比如可以运用multi-task learning的思想来解决这类问题。 Autonomy:本意是自治性。在这里表示FL system中的每个party都是under independent control的,每个party有权自己决定是否和其它party共享信息。 1) Association Autonomy:每个party有权决定随时加入或者退出FL system(比如设备没电了)。一个party甚至可以加入多个FL ...