Imbalanced data problem is widely present in network intrusion detection, spam filtering, biomedical engineering, finance, science, being a challenge in many real-life data-intensive applications. Classifier bias occurs when traditional classification algorithms are used to deal with imbalanced data. As already known, the General Vector Machine (GVM) algorithm has good generalization ability, though it does not work well for the imbalanced classification. Additionally, the state-of-the-art Binary Ant Lion Optimizer (BALO) algorithm has high exploitability and fast convergence rate. Based on these facts, we have proposed in this paper a Cost-sensitive Feature selection General Vector Machine (CFGVM) algorithm based on GVM and BALO algorithms to tackle the imbalanced classification problem, delivering different cost weights to different classes of samples. In our method, the BALO algorithm determines the cost weights and extract more significant features to improve the classification performance. Experiments conducted on eleven imbalanced data sets have shown that the CFGVM algorithm significantly improves the classification performance of minority class samples. By comparing with similar algorithms and state-of-the-art algorithms, the proposed algorithm significantly outperforms in performance and produces better classification results.

The top-k influential individuals in a social network under a specific topic play an important role in reality. Identifying top-k influential nodes of a social network is still an open and deeply-felt problem. In recent years, some researchers adopt the swarm intelligence algorithm to solve such problems and obtain competitive results. There are two main algorithm models for swarm intelligence, namely Ant Colony System (ACS) and Particle Swarm Optimization (PSO). The discretized basic Particle Swarm Algorithm (DPSO) shows comparable performance in identifying top-k influential nodes of a social network. However, the performance of the DPSO algorithm is directly related to the choice of its local search strategy. The local search strategy based on the greedy mechanism of the initial DPSO can easily lead to the global suboptimal solution due to the premature convergence of the algorithm. In this paper, we adopt the degree centrality based on different neighbourhoods to enhance its local search ability. Through experiments, we find that local search strategies based on different neighbourhoods have significant differences in the improvement of the algorithm's global exploration capabilities, and the enhancement of the DPSO algorithm based on the degree centrality of different neighbourhoods has a saturation effect. Finally, based on the degree centrality of the best neighbourhood with improved local search ability, we propose the DPSO_NDC algorithm. Experimental results in six real-world social networks show that the proposed algorithm outperforms the initial DPSO algorithm and other state-of-the-art algorithms in identifying the top-k influence nodes.

With the rapid development of microarray technology and interdisciplinary science, it is possible for microarray technology to be used to predict diseases. Microarray technology has the advantages of high speed, high efficiency and reliability in disease prediction. However, microarray data are usually high-dimensional with small samples, additionally, the samples are often imbalanced, which brings a lot of difficulties to researchers. In view of the above problems, it is proposed in this paper a Filter-Wrapper hybrid feature selection algorithm Union Information Gini Cost-sensitive Feature Selection General Vector Machine (UIG-CFGVM) to tackle the high-dimensional imbalanced small-sample problem. The improved hybrid algorithm is as follows: Firstly, the most common features are removed by the proposed hybrid filter algorithm UIG, which is obtained by Information Gain (Info)and Gini Index (Gini). Secondly, Cost-sensitive Feature selection General Vector Machine (CFGVM) is used as Wrapper method to further improve the performance of the algorithm. The experimental results show that the proposed algorithm UIG-CFGVM has better classification performance in seven biomedical high-dimensional imbalanced small-sample datasets compared with other similar algorithms.

During the beam current calibration process, accurate guidance of the beam current to the metal target is a challenging issue for proton accelerators. To address this challenge, we propose the use of beam orbital parameters combined with reinforcement learning algorithms to achieve automatic beam calibration. This study introduces a system architecture that employs edge intelligent acceleration nodes based on deep learning acceleration techniques. We designed a system to predict BPM parameters using a cascaded backpropagation neural network (CBPNN) that is informed by the physical structure. This system serves as an environmental map for reinforcement learning, aiding beam current correction. The CBPNN was implemented on the acceleration node to hasten the forward inference process, leveraging sparsification, quantization algorithms, and pipelining techniques. Our experimental results demonstrated that the simulated inference speed reached 28 μs with FPGA hardware as the edge acceleration node, achieving forward inference speeds 35.66 and 12.66 times faster than those of the CPU and GPU. The energy efficiency ratio was 10.582 MOPS/W, which was 989 and 410 times that of the CPU and GPU, respectively. This confirms the designed architecture’s energy efficiency and low latency attributes.

Accelerator-driven systems (ADS) are promising technologies for nuclear waste transmutation and energy production. The China ADS Front-end Superconducting Demo Linac (CAFe) is a prototype of the China Initiative Accelerator Driven System (CiADS), which aims to verify the feasibility of key technologies of CiADS. In this article, a novel method for historical data screening of the beam transport in the medium energy beam transport (MEBT) section of CAFe is presented. A clustering fusion algorithm based on unsupervised learning and beam transmission characteristics is designed to extract a large number of samples with the beam in steady-state transmission from historical beam data. A deep neural network model was constructed to fit the beam transport characteristics and verify the reliability of the screened data samples. The method can improve the efficiency and accuracy of data analysis and provide valuable insights for the optimization and control of beam transport in CAFe.