Multi-task learning (MTL) simultaneously learns multiple related tasks, aiming to improve overall outcomes by leveraging task correlations. The MTL network structure choice greatly impacts individual task results. In contrast, Evolutionary Neural Architecture Search (ENAS) has shown remarkable proficiency in automatically identifying optimal neural network structures for MTL. However, existing methods present certain limitations. Manually designed multi-task learning networks are constrained by the designer’s skill, and they often fail to fully capture multi-task feature interactions within the backbone network. Moreover, these networks do not incorporate the residual concept to preserve crucial backbone network output features. In contrast, automatically searched structures embrace residuals but still need improvement in facilitating multi-task feature interactions. This paper introduces a novel framework for evolutionary MTL network search. It includes a backbone with residuals and a Multi-Task Feature Fusion and Extraction Module (FFEM) with channel attention. The FFEM combines task features and weighs channel(feature) importance. Adopting the residual concept reintegrates them into the backbone network and fuses with the backbone network’s features. Experimental results validate our approach’s efficacy, achieving competitive performance on the NYUD-v2 dataset.

In recent years, an increasing number of map apps have provided route planning services to their users. However, the quality of route planning services relies heavily on having correct data about transportation infrastructure. As many planned subway lines are being built across cities, there are conflicts between the actual conditions and the data provided by map apps for temporary bus stops, which may result in complaints against public transportation operators. However, it is difficult to tackle these complaints, as public transportation operators can obtain only inaccurate information about the locations of temporary bus stops. To resolve these conflicts, crowd bus sensing (CBS) is proposed in this paper. CBS is a new sensing paradigm that takes advantage of the extensive deployment of GPS trackers and prior knowledge about the transportation infrastructures covered by scheduled bus routes. Extensive experimental evaluations on real-world and synthetic datasets show that the proposed CBS system outperforms state-of-the-art methods.

Diagnosis of skin lesions based on imaging techniques remains a challenging task because data (knowledge) uncertainty may reduce accuracy and lead to imprecise results. This paper investigates a new deep hyperspherical clustering (DHC) method for skin lesion medical image segmentation by combining deep convolutional neural networks and the theory of belief functions (TBF). The proposed DHC aims to eliminate the dependence on labeled data, improve segmentation performance, and characterize the imprecision caused by data (knowledge) uncertainty. First, the SLIC superpixel algorithm is employed to group the image into multiple meaningful superpixels, aiming to maximize the use of context without destroying the boundary information. Second, an autoencoder network is designed to transform the superpixels' information into potential features. Third, a hypersphere loss is developed to train the autoencoder network. The loss is defined to map the input to a pair of hyperspheres so that the network can perceive tiny differences. Finally, the result is redistributed to characterize the imprecision caused by data (knowledge) uncertainty based on the TBF. The proposed DHC method can well characterize the imprecision between skin lesions and non-lesions, which is particularly important for the medical procedures. A series of experiments on four dermoscopic benchmark datasets demonstrate that the proposed DHC yields better segmentation performance, increasing the accuracy of the predictions while can perceive imprecise regions compared to other typical methods.

Knowledge bases (KBs) provide a large amount of structured information for entities and relations, which are successfully leveraged in many natural language processing tasks. However, distantly supervised relation extraction only utilizes KBs to automatically generate datasets, while ignoring the b ackground information in KBs during the relation extraction process. We herein propose a knowledge-embodied attention that leverages knowledge information in KBs to reduce the impact of noisy data for distantly supervised relation extraction. Specifically, we pre-train distributed representations of KBs with the knowledge representation learning (KRL) model, and subsequently incorporate them into relation extraction to learn sentence-level attention weights. The experimental results demonstrate that our approach outperforms all baselines, thus indicating that we can focus our attention on valid data by leveraging background information in KBs.

In the era of big data, the data provided by multiple sources for the same entity may result in conflicting information. Because sources with higher reliability degrees provide true information more frequently, the truth can be obtained by estimating the weights or reliability degrees of the source s. Due to hostile websites or faulty sensors, some sources may occasionally provide outliers that deviate significantly from the truth. However, in the majority of existing truth discovery methods, each source should be uniformly assigned a weight at the initial stage. Therefore, the accuracy of truth estimation is degraded by outliers. Several previous studies have proposed kernel density estimation-based truth discovery algorithms to solve the problems caused by outliers. These approaches aim to estimate the probability distribution of observation values to assess the reliability of sources. Unfortunately, the data with outliers that are smoothed by Gaussian kernels may cause more deviations from the truth. Thus, we propose a local linear regression (LLR) method for addressing the problems caused by outliers. The proposed method can effectively estimate the source reliability and the truth of the datasets with outliers. Experiments on two real-world datasets demonstrate that the proposed method yields more accurate results than existing state-of-the-art methods.

An increasing number of studies have indicated that long-non-coding RNAs (lncRNAs) play critical roles in many important biological processes. Predicting potential lncRNA-disease associations can improve our understanding of the molecular mechanisms of human diseases and aid in finding biomarkers for disease diagnosis, treatment, and prevention. In this paper, we constructed a bipartite network based on known lncRNA-disease associations; based on this work, we proposed a novel model for inferring potential lncRNA-disease associations. Specifically, we analyzed the properties of the bipartite network and found that it closely followed a power-law distribution. Moreover, to evaluate the performance of our model, a leave-one-out cross-validation (LOOCV) framework was implemented, and the simulation results showed that our computational model significantly outperformed previous state-of-the-art models, with AUCs of 0.8825, 0.9004, and 0.9292 for known lncRNA-disease associations obtained from the LncRNADisease database, Lnc2Cancer database, and MNDR database, respectively. Thus, our approach may be an excellent addition to the biomedical research field in the future.