Federated learning is a promising framework to train neural networks with widely distributed data. However, performance degrades heavily with heterogeneously distributed data. Recent work has shown this is due to the final layer of the network being most prone to local bias, some finding success freezing the final layer as an orthogonal classifier. We investigate the training dynamics of the classifier by applying SVD to the weights motivated by the observation that freezing weights results in constant singular values. We find that there are differences when training in IID and non-IID settings. Based on this finding, we introduce two regularization terms for local training to continuously emulate IID settings: (1) variance in the dimension-wise probability distribution of the classifier and (2) hyperspherical uniformity of representations of the encoder. These regularizations promote local models to act as if it were in an IID setting regardless of the local data distribution, thus off-setting proneness to bias while being flexible to the data. On extensive experiments in both label-shift and feature-shift settings, we verify that our method achieves highest performance by a large margin especially in highly non-IID cases in addition to being scalable to larger models and datasets.

Abstract As wireless mobile network technologies have grown tremendously over the past few years, security issues of protecting user privacy and network resources have emerged. Particularly, a mobile node (MN) must be authenticated in order to utilize the network resources whenever the MN performs its initial access or handover. In this paper, we present secure handover scenarios for Proxy Mobile IPv6 (PMIPv6): namely, handover re‐authentication (HORA)and handover early‐authentication (HOEA). The presented secure handover scenarios are based on EAP that is a common authentication framework for wireless mobile networks. Analysis results for the total handover latency and packet loss are presented to demonstrate whether the presented secure handover scenarios satisfy with real‐time applications. PMIPv6 is in the early stage of deployment in various wireless mobile networks. The conducted results in this paper thus offer the key to an understanding of feasibility of PMIPv6 in the coming next‐generation network environments. Copyright © 2009 John Wiley & Sons, Ltd.

Proxy Mobile IPv6 has been developed from the concept of network-based mobility support protocol in the IETF. The recently published specification of proxy mobile IPv6 only focuses on the mobility support. Then, route optimization issues are left in the basket for further works. In this paper, we provide the performance analysis in where the recently proposed route optimization is evaluated in terms of signaling cost and packet delivery cost. Through the provided analysis results, we demonstrate that route optimization solves the ineffective routing path problem during the mobile host communicates with its corresponding host and argue that the scalability of proxy mobile IPv6 architecture is improved due to the distributed routing path. In addition, the cost model developed in this paper would be a reference model in order to facilitate decision-making for further route optimization design.

In recent years, high-bandwidth and low-cost wireless technologies have emerged as a competitive element, enabling the smart home environment. On the other hand, with the increasing demands for various services, the current access point allocation schemes give rise to significant challenges for a stable connection service, bandwidth and load distribution. In this paper, we present an optimal access point allocation scheme based on genetic algorithm that attempts to optimize multiple parameters, such as bandwidth, and load-balancing requirements. The proposed allocation scheme provides a set of approximately efficient solutions, which allows a wireless user to choose an access point based on its capacity and load balancing, not only regarding its signaling strength or service set identifier. The simulation results are provided to demonstrate the impact of the proposed optimization procedure on overall system performance in terms of connection distribution, load balancing and call dropping probability.

Mobile IP is the basic solution to providing host mobility, whereas network mobility (NEMO) refers to the concept of the collective mobility of a set of nodes. The NEMO basic support protocol has been proposed in IETF as a first solution to the problem of network mobility. The main limitation of this basic solution is that it forces triangular routing, i.e., packets are always forwarded through the home agent (HA), following a suboptimal path. This is because each sub-NEMO obtains a care of address (CoA) that belongs to the home prefix of its parent mobile router. Such a CoA is not topologically meaningful in the current location, since the parent mobile router could also be away from home, and hence, packets addressed to the CoA are forwarded through the HA of the parent NEMO. To solve this problem, various extended proposals, with differing approaches and goals, exist for route optimization (RO) in NEMO applications. Their influences on the RO performance have been evaluated by classifying the detailed operations performed within the nested NEMO network, and then each category is analyzed in detail. The modeling of the detailed RO operation is intended to quantify the tradeoffs between the different approaches in order to provide a basis for the selection decision. In particular, the proposed grouping of the different proposals, based on their address configuration strategy, clarifies their similarities and differences, and provides some useful insights into the various methods that have been developed. In conclusion, it is suggested that, when choosing a solution for deploying NEMO, the designer has to balance his choices between the different pros and cons, and the different cases of application that are derived in this paper.

Proxy mobile IPv6 is one of network-based mobility support protocols which does not require a mobile host to be involved in mobility support signaling. Even though proxy mobile IPv6 has many advantages than client-based mobility support protocols such as mobile IPv6, unfortunately, it may cause an out-of-sequence problem during its route optimization procedure. Thus, in this paper, we develop an estimation function for an out- of-sequence time period and reveal the out-of-sequence problem caused by the route optimization procedure. Then we propose a scheme which minimizes the arrival of out-of-sequence packets to the mobile host and analyze the impact of out-of-sequence packets. The performance evaluation results demonstrate that the traffic rate and out-of-sequence time period mainly have an effect on the out-of-sequence problem.

In this paper, we present a comparative signaling cost analysis of hierarchical mobile IPv6 and proxy mobile IPv6. The two mobility support protocols have a similar hierarchical mobility management architecture. They, however, are clearly different: Hierarchical mobile IPv6 has a specific properties of host-based mobility support protocol and proxy mobile IPv6 is a kind of network-based mobility support protocols. Thus, it is important to reveal their mobility characters and performance impact factors. In this paper, we develop a cost model that evaluates the location update cost and the packet delivery cost based on protocol operations. Then, we present the numerical results in where the impacts of various system parameters are evaluated.

This paper presents a policy-based hybrid management architecture for IP-based VPN (virtual private network) service that makes use of a centralized global policy management server.