Search

Clustering of cell types from a large number of high-dimensional heterogeneous cells is a vital step in analyzing single-cell RNA-seq data. Although several computational methods have been proposed to evolve such data, most of them suffer from some limitations such as high-level noise, high dimensionality, and low generalization. To address these challenges, a multiobjective robust continuous clustering algorithm (MORCC) is presented to discriminate the different cell types in a single-cell RNA-seq dataset. Stepwise, first a dimensionality reduction method is applied to map the high-dimensional heterogeneous cells into a desired low-dimensional space while preserving the features of the original space. Then, to overcome the instability of trial-and-error connectivity weights in the robust continuous clustering, MORCC proposes applying evolutionary operators to optimize the connectivity weights dynamically, and to select the suitable parameters with two cluster validity indices. To demonstrate the effectiveness of MORCC, we compare it to several state-of-the-art methods on six single-cell RNA-seq datasets, revealing its superior clustering ability from several perspectives. In addition, we carry out a parameter analysis, a case study, and visualization and biological interpretability analyses to validate MORCC's cell identification capability on single-cell RNA-seq data. [ABSTRACT FROM AUTHOR]

Deep convolutional neural networks have been widely used in numerous applications, but their demanding storage and computational resource requirements prevent their applications on mobile devices. Knowledge distillation aims to optimize a portable student network by taking the knowledge from a well-trained heavy teacher network. Traditional teacher–student-based methods used to rely on additional fully connected layers to bridge intermediate layers of teacher and student networks, which brings in a large number of auxiliary parameters. In contrast, this article aims to propagate information from teacher to student without introducing new variables that need to be optimized. We regard the teacher–student paradigm from a new perspective of feature embedding. By introducing the locality preserving loss, the student network is encouraged to generate the low-dimensional features that could inherit intrinsic properties of their corresponding high-dimensional features from the teacher network. The resulting portable network, thus, can naturally maintain the performance as that of the teacher network. Theoretical analysis is provided to justify the lower computation complexity of the proposed method. Experiments on benchmark data sets and well-trained networks suggest that the proposed algorithm is superior to state-of-the-art teacher–student learning methods in terms of computational and storage complexity. [ABSTRACT FROM AUTHOR]

Stratifying personalized treatment for patients has been one of the main challenges for modern medicine. To solve this problem, various clustering algorithms have been proposed for patient stratification in both quantification and biological ways meaningfully. However, most of the existing clustering algorithms still suffer from many realistic algorithm limitations such as low diagnostic ability and bad generalization. Therefore, to address those restrictions, we propose a novel multiobjective spectral clustering algorithm based on decomposition. A population that consists of distance weight and two other indispensable parameters of the spectral clustering is optimized by the proposed algorithm. Two cluster validity indices are proposed to capture the characteristics of different datasets. To validate the effectiveness and efficiency of the proposed algorithm, we benchmark it on thirty-five real patient stratification datasets and six real-world medical datasets across thousands of comparisons with fifteen algorithms, including ten effective clustering methods and five state-of-the-art multiobjective algorithms. The experimental results indicate that the proposed algorithm performs better than other compared algorithms with high clustering ability for patient stratification. Moreover, extensive analysis of time complexity and parameters are performed to prove the robustness of the proposed algorithm from different perspectives. [ABSTRACT FROM AUTHOR]

Unmanned aerial vehicles (UAVs) searching for a moving target is a complex search optimization problem that aims to find a path with maximum probability of observing the target. However, as the area where the target may be located increases in size, existing methods tend to cause easy entrapment into local optima, resulting in poor performance. Hence, we propose a Convex Combination Multiple Populations Competitive Swarm Optimization algorithm (CDCSO) in this paper. First, we suggest applying the multiple population strategy to enhance the search ability of the algorithm. The population of interest is divided into two subgroups that each conduct the CSO. Second, we propose a novel convex combination update strategy to coordinate the two subgroups to search jointly for the global optimal solution. At each iteration, this strategy allows the two subgroups to compete and learn from each other, which prevents the population from falling into the local optimum. We then designed new scenarios containing 3 to 6 probability areas to investigate the performance of our proposed algorithm in multiple and complex scenarios. Results: To demonstrate the effectiveness and stability of the proposed algorithm, we compared it with several swarm optimization algorithms from different perspectives. Experimental results showed that the proposed algorithm has superior performance in following and intercepting a high-probability area. To further demonstrate the generality of the convex combination update strategy, we evaluated the strategy in several developed particle swarm optimization algorithms. The results demonstrated that the convex combination update strategy can be integrated into other algorithms to better address the moving target search problem. [ABSTRACT FROM AUTHOR]