Search

Federated Learning (FL) enables the multiple participating devices to collaboratively contribute to a global neural network model while keeping the training data locally. Unlike the centralized training setting, the non-IID, imbalanced (statistical heterogeneity) and distribution shifted training data of FL is distributed in the federated network, which will increase the divergences between the local models and the global model, further degrading performance. In this paper, we propose a flexible clustered federated learning (CFL) framework named FlexCFL, in which we 1) group the training of clients based on the similarities between the clients' optimization directions for lower training divergence; 2) implement an efficient newcomer device cold start mechanism for framework scalability and practicality; 3) flexibly migrate clients to meet the challenge of client-level data distribution shift. FlexCFL can achieve improvements by dividing joint optimization into groups of sub-optimization and can strike a balance between accuracy and communication efficiency in the distribution shift environment. The convergence and complexity are analyzed to demonstrate the efficiency of FlexCFL. We also evaluate FlexCFL on several open datasets and made comparisons with related CFL frameworks. The results show that FlexCFL can significantly improve absolute test accuracy by +10.6% on FEMNIST compared to FedAvg, +3.5% on FashionMNIST compared to FedProx, +8.4% on MNIST compared to FeSEM. The experiment results show that FlexCFL is also communication efficient in the distribution shift environment., Comment: Manuscript under review. arXiv admin note: substantial text overlap with arXiv:2010.06870

Wireless blockchain network is proposed to enable a decentralized and safe wireless networks for various blockchain applications. To achieve blockchain consensus in wireless network, one of the important steps is to broadcast new block using wireless channel. Under wireless network protocols, the block transmitting will be affected significantly. In this work, we focus on the consensus process in blockchain-based wireless local area network (B-WLAN) by investigating the impact of the media access control (MAC) protocol, CSMA/CA. With the randomness of the backoff counter in CSMA/CA, it is possible for latter blocks to catch up or outpace the earlier one, which complicates blockchain forking problem. In view of this, we propose mining strategies to pause mining for reducing the forking probability, and a discard strategy to remove the forking blocks that already exist in CSMA/CA backoff procedure. Based on the proposed strategies, we design Block Access Control (BAC) approaches to effectively schedule block mining and transmitting for improving the performance of B-WLAN. Then, Markov chain models are presented to conduct performance analysis in B-WLAN. The results show that BAC approaches can help the network to achieve a high transaction throughput while improving block utilization and saving computational power. Meanwhile, the trade-off between transaction throughput and block utilization is demonstrated, which can act as a guidance for practical deployment of blockchain.

This work proposes a thermophysical model for realistic surface layers on airless small bodies (RSTPM), for the use of interpreting their multi-epoch thermal lightcurves (e.g WISE/NEOWISE). RSTPM considers real orbital cycle, rotation cycle, rough surface, temperature dependent thermal parameters, as well as contributions of sunlight reflection to observations, hence being able to produce precise temperature distribution and thermal emission of airless small bodies regarding the variations in orbital time scales. Details of the physics, mathematics and numerical algorithms of RSTPM are presented. When used to interpret multi-epoch thermal lightcurves by WISE/NEOWISE, RSTPM can give constraints on the spin orientation and surface physical properties, like mean thermal inertia or mean size of dust grains, roughness fraction, albedo and so on via radiometric procedure. As an application example, we apply this model to the main-belt object (24) Themis, the largest object of the Themis family, which is believed to be the source region of many main-belt comets. We find multi-epoch (2010, 2014-2018) observations of Themis by WISE/NEOWISE, yielding 18 thermal lightcurves. By fitting these data with RSTPM, best-fit spin orientation of Themis is derived to be ($\lambda=137^\circ$, $\beta=59^\circ$) in ecliptic coordinates, the mean radius of dust grains on the surface is estimated to be $\tilde{b}=140^{+500}_{-114}(6\sim640)~\mu$m, indicating the surface thermal inertia to vary from $\sim3\rm~Jm^{-2}s^{-0.5}K^{-1}$ to $\sim60\rm~Jm^{-2}s^{-0.5}K^{-1}$ due to seasonal temperature variation. Further analysis found that thermal light curves of Themis show a weak rotation-phase dependent feature, indicative of heterogeneous thermal properties or imperfections of lightcurve inversion shape model., Comment: 23 pages, 22 figures, ApJ in press

Machine learning technologies using deep neural networks (DNNs), especially convolutional neural networks (CNNs), have made automated, accurate, and fast medical image analysis a reality for many applications, and some DNN-based medical image analysis systems have even been FDA-cleared. Despite the progress, challenges remain to build DNNs as reliable as human expert doctors. It is known that DNN classifiers may not be robust to noises: by adding a small amount of noise to an input image, a DNN classifier may make a wrong classification of the noisy image (i.e., in-distribution adversarial sample), whereas it makes the right classification of the clean image. Another issue is caused by out-of-distribution samples that are not similar to any sample in the training set. Given such a sample as input, the output of a DNN will become meaningless. In this study, we investigated the in-distribution (IND) and out-of-distribution (OOD) adversarial robustness of a representative CNN for lumbar disk shape reconstruction from spine MR images. To study the relationship between dataset size and robustness to IND adversarial attacks, we used a data augmentation method to create training sets with different levels of shape variations. We utilized the PGD-based algorithm for IND adversarial attacks and extended it for OOD adversarial attacks to generate OOD adversarial samples for model testing. The results show that IND adversarial training can improve the CNN robustness to IND adversarial attacks, and larger training datasets may lead to higher IND robustness. However, it is still a challenge to defend against OOD adversarial attacks., Comment: Published at SPIE Medical Imaging: Image Processing 2021