Machine learning algorithms are commonly being deployed in decision-making systems that have a direct impact on human lives. However, if these algorithms are trained solely to minimize training/test errors, they may inadvertently discriminate against individuals based on their sensitive attributes, such as gender, race or age. Recently, algorithms that ensure the fairness are developed in the machine learning community. Fairness criteria are applied by these algorithms to measure the fairness, but they often use the point estimate to assess the fairness and fail to consider the uncertainty of the sample fairness criterion once the algorithms are deployed. We suggest that assessing the fairness should take the uncertainty into account. In this paper, we use the covariance as a proxy for the fairness and develop the confidence region of the covariance vector using empirical likelihood \citep{Owen1988}. Our confidence region based fairness constraints for classification take uncertainty into consideration during fairness assessment. The proposed confidence region can be used to test the fairness and impose fairness constraint using the significant level as a tool to balance the accuracy and fairness. Simulation studies show that our method exactly covers the target Type I error rate and effectively balances the trade-off between accuracy and fairness. Finally, we conduct data analysis to demonstrate the effectiveness of our method.

Disruptive network communication entails transient network connectivity, asymmetric links, and unstable nodes, which pose severe challenges to data collection in sensor networks. Erasure coding can be applied to mitigate the dependency of feedback in such a disruptive network condition, improving data collection. However, the collaborative data collection through an in-network erasure coding approach has been underexplored. In this paper, we present an Opportunistic Network Erasure Coding protocol (ONEC) to collaboratively collect data in dynamic disruptive networks. ONEC derives the probability distribution of coding degree in each node and enables opportunistic in-network recoding, and guarantees that the recovery of original sensor data can be achieved with high probability upon receiving any sufficient amount of encoded packets. First, it develops a recursive decomposition structure to conduct probability distribution deconvolution, supporting heterogeneous data rates. Second, every node conducts selective in-network recoding of its own sensing data and received packets, including those opportunistic overheard packets. Last, ONEC can efficiently recover raw data from received encoded packets, taking advantages of low decoding complexity of erasure codes. We evaluate and show that our ONEC can achieve efficient data collection in various disruptive network settings. Moreover, ONEC outperforms other epidemic network coding approaches in terms of network goodput, communication cost, and energy consumption.

Sensor network has found critical applications in extreme environments. However, in the extreme environments, a predictable and stable path may never exist, since the transient network connectivity, asymmetric links and unstable nodes are prevalent. Thus, the extreme environments severely challenge its basic function of data collection. Particularly, those disruptive conditions make traditional data collection protocols inefficient. In this paper, we design Opportunistic Network Erasure Coding (ONEC) protocol for collaborative data collection in disruptive sensor networks. The idea behind ONEC protocol is to study how each node determines code degree distribution and receding strategies in a distributed fashion, so that all data of network can be recovered with high probability upon receiving any sufficient amount of encoded packets. First, each node derives code degree distribution through recursive discrete deconvolution. Second, every node conducts selective receding of its own sensing data and opportunistically received data. Last, the ONEC ensures decoder can recover all data from any sufficient amount of encoded packets with high probability. The performance evaluations through extensive simulation validate that ONEC can truly achieve efficient data collection with high reliability in disruptive sensor networks and outperform other existing approaches in terms of network goodput, message complexity and buffer space.

A typical microarray gene expression dataset is usually both extremely sparse and imbalanced. To select multiple highly informative gene subsets for cancer classification and diagnosis, a new Fuzzy Granular Support Vector Machine---Recursive Feature Elimination algorithm (FGSVM-RFE) is designed in this paper. As a hybrid algorithm of statistical learning, fuzzy clustering, and granular computing, the FGSVM-RFE separately eliminates irrelevant, redundant, or noisy genes in different granules at different stages and selects highly informative genes with potentially different biological functions in balance. Empirical studies on three public datasets demonstrate that the FGSVM-RFE outperforms state-of-the-art approaches. Moreover, the FGSVM-RFE can extract multiple gene subsets on each of which a classifier can be modeled with 100% accuracy. Specifically, the independent testing accuracy for the prostate cancer dataset is significantly improved. The previous best result is 86% with 16 genes and our best result is 100% with only eight genes. The identified genes are annotated by Onto-Express to be biologically meaningful.