In this paper we are interested in what we term the federated private bandits framework, that combines differential privacy with multi-agent bandit learning. We explore how differential privacy based Upper Confidence Bound (UCB) methods can be applied to multi-agent environments, and in particular to federated learning environments both in ‘master-worker’ and ‘fully decentralized’ settings. We provide a theoretical analysis on the privacy and regret performance of the proposed methods and explore the tradeoffs between these two.
We study federated learning (FL) at the wireless edge, where power-limited devices with local datasets train a joint model with the help of a remote parameter server (PS). We assume that the devices are connected to the PS through a bandwidth-limited shared wireless channel. At each iteration of FL, a subset of the devices are scheduled to transmit their local model updates to the PS over orthogonal channel resources. We design novel scheduling policies, that decide on the subset of devices to transmit at each round not only based on their channel conditions, but also on the significance of their local model updates. Numerical results show that the proposed scheduling policy provides a better long-term performance than scheduling policies based only on either of the two metrics individually. We also observe that when the data is independent and identically distributed (i.i.d.) across devices, selecting a single device at each round provides the best performance, while when the data distribution is non-i.i.d., more devices should be scheduled.
In this paper, we study the problem of federated learning (FL) over a wireless channel, modeled by a Gaussian multiple access channel (MAC), subject to local differential privacy (LDP) constraints. We show that the superposition nature of the wireless channel provides a dual benefit of bandwidth efficient gradient aggregation, in conjunction with strong LDP guarantees for the users. We propose a private wireless gradient aggregation scheme, which shows that when aggregating gradients from K users, the privacy leakage per user scales as O(1√K) compared to orthogonal transmission in which the privacy leakage scales as a constant. We also present analysis for the convergence rate of the proposed private FL aggregation algorithm and study the tradeoffs between wireless resources, convergence, and privacy.
Federated learning (FL) refers to the adaptation of a central model based on data sets available at multiple remote users. Two of the common challenges encountered in FL are the fact that training sets obtained by different users are commonly heterogeneous, i.e., arise from different sample distributions, and the need to communicate large amounts of data between the users and the central server over the typically expensive uplink channel. In this work we formulate the problem of FL in which different clusters of users observe labeled samples drawn from different distributions, while operating under constraints on the communication overhead. For such settings, we identify that the combination of statistical heterogeneity and communication constraints induces an inherent tradeoff between the ability of the users of each cluster to learn a proper model and the accuracy in aggregating these models into a global inference rule. We propose an algorithm based on multi-source adaptation methods for such communication-aware clustered FL scenarios which allows to balance these performance measures, and demonstrate its ability to achieve improved inference over conventional federated averaging without inducing additional communication overhead.