Deep learning as a service (DLaaS) that promotes deep learning-based applications by selling computing services from IT companies to end-users has introduced potential privacy leaks from users and cloud servers. Federated learning (FL) provides an emerging distributed paradigm that enables numerous users to collaboratively train deep-learning models while protecting user privacy and data security. However, many FL-related existing works only focus on improving communication bottlenecks due to frequent model parameter transmission, but ignore the performance degradation incurred by imbalanced user distribution and high inference latency due to the high complexity of deep-learning models in the emerging IoT-edge-cloud FL. In this paper, we propose an efficient user-distribution-aware hierarchical FL for communication-efficient training and fast inference in the IoT-edge-cloud DLaaS architecture. Specifically, we propose a user-distribution-aware hierarchical FL architecture to cope with the performance degradation owing to the imbalanced user distribution. The proposed architecture also features a lightweight deep neural network that adopts the designed lightweight fire modules as components and has a side branch for communication-efficient training and fast inference. Extensive experiments demonstrate that the proposed schemes significantly boost the accuracy by up to 67.12%, save 47.98% communication costs, and accelerate inference by up to 87.24$\boldsymbol{\times}$compared to benchmarking methods.

Survival analysis with high-dimensional data deals with the prediction of patient survival based on their gene expression data and clinical data. A crucial task for the accuracy of survival analysis in this context is to select the features highly correlated with the patient's survival time. Since the information about class labels is hidden, existing feature selection methods in machine learning are not applicable. In contrast to classical statistical methods which address this issue with the Cox score, we propose to tackle this problem by discretizing the survival time of patients into a suitable number of subgroups via silhouettes clustering validity. To cope with patients' censoring, we use “k-nearest neighbor” based on clinical parameters. Feature selection is then accomplished using Fast Correlation-Based Filtering approach from machine learning community. The effectiveness and efficiency of the proposed method are demonstrated through comparisons with classical statistical methods on real-world datasets and simulation datasets.

Importance sampling is used in this paper to address the classical yet important problem of performance estimation of block codes. Simulation distributions that comprise discrete- and continuous-mixture probability densities are motivated and used for this application. These mixtures are employed in concert with the so-called g-method, which is a conditional importance sampling technique that more effectively exploits knowledge of underlying input distributions. For performance estimation, the emphasis is on bit by bit maximum a-posteriori probability decoding, but message passing algorithms for certain codes have also been investigated. Considered here are single parity check codes, multidimensional product codes, and briefly, low-density parity-check codes. Several error rate results are presented for these various codes, together with performances of the simulation techniques.