Dynamic Constrained Multiobjective Optimization Problems (DCMOPs) are very difficult to solve because both of the objectives and constraints may change over time. The existing approaches for solving DCMOPs mainly develop dynamic response techniques and constraint handling techniques. But they do not focus on the search capability of the static optimizer in each environment, which ignores the intrinsic requirement of quickly locating Pareto-optimal Front (PF) in each environment when solving DCMOPs. To this end, this paper proposes a dual-population evolutionary algorithm for solving DCMOPs, called as DpEA, which maintains a population without considering constraints (called UP) for exploration and a population with considering constraints (called CP) for exploitation in each environment. In each iteration of a new environment, UP firstly adopts a stratified mutation strategy (SMS) and a dominated solution repairment strategy (DSR) to enhance the exploration ability of finding promising regions where the PF may reside. SMS uses solutions from different nondominated fronts to generate offspring, while DSR repairs the single-optimal variables of the dominated solutions by sampling from the distribution of those variables of nondominated solutions. Secondly, this paper uses an adaptive offspring ratio adjustment strategy to control the offspring number generated by UP and CP according to the normalized Hausdorff distance between nondominated solution sets from the two latest generations of UP. This strategy is helpful to balance the intensity between exploration and exploitation and thereby ensures efficient search. Experimental results on CEC 2023 DCF test suite show that DpEA has a superior performance over six state-of-the-art algorithms.

For the latest two years, relation classification-based surrogate-assisted algorithms show good potential for solving expensive multi-objective optimization problems (EMOPs). In this category of methods, the used dominance relation that is vital for building training dataset and selecting promising solutions to reduce expensive real function evaluations (FEs). However, the existing studies are still at the initial stage and lack specific research on the dominance relation. This paper proposes a novel dominance relation called Difference Vector Angle Dominance with an angle threshold for EMOPs (called as DVAD- φ ). The proposed DVAD- φ has adaptive selection pressure and considers the convergence and diversity of solutions when picking out superior solutions, which makes it beneficial to pick out promising solutions for expensive real FEs and reduce expensive real FEs. To be specific, we firstly give the definition of DVAD- φ that measures the superiority from one solution to another solution, where the angle threshold φ controls the selection pressure. Then, we propose an adaptive determination strategy of angle threshold based on bisection to set proper pressure for picking out promising solutions for expensive real FEs. Experiments have been conducted on 7 test functions from one benchmark set. The experimental results have verified the effectiveness of DVAD- φ .

Meta-heuristic algorithms are popular for their efficiency in solving complex optimization problems. Although there are many known algorithms, identifying ways to improve their performance remains an important research area. This paper proposes a brain neuroscience-inspired meta-heuristic algorithm called the Neural Population Dynamics Optimization Algorithm (NPDOA). There are three strategies in NPDOA. (1) The attractor trending strategy drives neural populations towards optimal decisions, thereby ensuring exploitation capability. (2) The coupling disturbance strategy deviates neural populations from attractors by coupling with other neural populations, thus improving exploration ability. (3) The information projection strategy controls the communication between neural populations, enabling a transition from exploration to exploitation. The results of benchmark and practical problems verified the effectiveness of NPDOA.

The existing functional module detection methods of Protein-Protein Interaction Network(PPIN), which are based on network embedding, usually only embed the information of protein nodes and do not embed the information of edge weights between proteins, which deteriorates the quality of protein functional module detection.To solve this problem, a functional module detection method based on deep network embedding of edge weighing information in PPIN is proposed. Combined with the topological structure of PPIN and attribute information of Gene Ontology(GO), the first-order edge weight information between proteins is measured using the attention coefficient of Graph ATtention(GAT) network, and the first-order edge weight information of proteins is embedded based on neighborhood aggregation.The forget and input gates of a Long Short-Term Memory(LSTM) network are used to measure the high-order edge weight information between proteins, whereby this information is embedded.According to the low-dimensional vector obtained by network embedding, the core clique is mined by the core attachment clustering algorithm, and the affiliate proteins are added to obtain the final protein functional module. Experimental results on the Collins, Gavin, and Krogan datasets show that the proposed method improves the accuracy and F1 score by up to 18.1 and 12.9 percentage points, respectively, compared with the methods such as COACH.

Charging infrastructure planning (CIPL) is key to popularizing electric vehicles and reducing carbon emissions. CIPL consists of two subproblems: charging station siting and charging pile allocation. The existing methods independently solve the two subproblems and ignore their interaction, which restricts the rationality of CIPL. To address this issue, this paper proposes a dual ant colony optimization for CIPL (DACO-CIPL). In each iteration, under the guidance of heuristic information and pheromones, the upper and lower ant colonies construct solutions for charging station siting and charging pile allocation in turn, respectively. Then, a global pheromone update strategy is performed to update the pheromones of each ant colony according to the historical best solutions, which realizes information transmission from the lower ant colony to the upper ant colony. In addition, whenever the upper ant colony finishes constructing solutions, a pheromone enhancement strategy is used to strengthen the pheromones of the lower ant colony according to the solutions of the upper ant colony, which realizes information transmission from the upper ant colony to the lower ant colony. DACO-CIPL is compared with several algorithms on multiple test instances. The experimental results show that DACO-CIPL has superior performance and more reasonable options for CIPL.

Discriminant characteristics of brain functional connectivity can be used as a biomarker for the diagnosis of neuropsychiatric diseases. Using a machine learning method to identify is an important topic in brain science research.Most of the existing recognition methods of brain functional connectivity biomarkers ignore the impact of the characteristics of high-dimensional, continuous and multi noise of brain functional connectivity data on the recognition performance, resulting in the weak classification ability of the obtained biomarkers. This paper proposes a brain function connectivity biomarker recognition method based on neighborhood decision rough sets. Firstly, according to the characteristics of continuity and high noise of brain function connectivity data, a neighborhood decision rough set,which can effectively deal with continuous and high noise data, is introduced to identify brain functional connectivity discriminant features with stronger classification ability as biomarkers. Then, according to the high-dimensional characteristics of brain function connectivity data, the efficiency of identifying brain function connectivity biomarkers by the neighborhood decision rough set is guaranteed by quickly generating the neighborhood and reducing the feature search space. The experimental results on the ABIDE I and ADNI data sets show that the proposed method can accurately and quickly obtain the discriminant features of brain functional connectivity with stronger classification ability, and is expected to provide more accurate biomarkers for the diagnosis of neuropsychiatric diseases.

Compared to general multi-objective optimization problems, multimodal multi-objective optimization problems (MMOPs) with local Pareto sets (PSs) must determine multiple global and local PSs simultaneously. Therefore, MMOPs with local PSs are challenging. To resolve this issue, this study proposes a multimodal multi-objective optimization evolutionary algorithm based on two-stage species conservation (MMOEA/TSC). MMOEA/TSC divides the evolutionary process into two stages: diversity-oriented species conservation and convergence-oriented species conservation. The former is aimed at locating promising regions in which global and local PSs may exist. To balance the distribution of solutions, a Gaussian variation strategy is used to iteratively generate diverse offspring in regions that contain the smallest number of solutions. The latter mainly focused on obtaining one PS with good convergence in each promising region. To help the solutions converge to the global and local PSs uniformly, a species stratification strategy was adopted according to the Pareto level of the well-converged solution for each species. The proposed algorithm was compared with seven state-of-the-art algorithms. For the CEC 2020 MMOP test problem set, the experimental results show that MMOEA/TSC has the capacity to find global and local PSs.

Despite that evolutionary and swarm intelligence algorithms have achieved considerable success on multi-objective optimization problems, they face huge challenges when dealing with many-objective optimization problems (MaOPs). There is an urgent call for effective evolutionary and swarm intelligence algorithms for MaOPs. Inspired by the satisfactory performance of bacterial foraging optimization (BFO) on the single-objective optimization problems, this paper extends BFO to deal with MaOPs and proposes a knowledge guided BFO for MaOPs (called as KLBFO). Firstly, KLBFO learns promising direction knowledge based on group decision making idea to guide the population to converge toward proper directions. Secondly, KLBFO learns elite knowledge by a new biological mechanism to accelerate the population to converge. Thirdly, KLBFO learns density knowledge by a new diversity management strategy based on orthogonal grid to produce well-distributed solutions. The performance of KLBFO is comprehensively evaluated by comparing it with eight state-of-the-art algorithms on two suites of test problems and one real-world problem. The empirical results have validated the superior performance of KLBFO for MaOPs.

Classification of dynamic functional connectivity (DFC) is becoming a promising approach for diagnosing various neurodegenerative diseases. However, the existing methods generally face the problem of overfitting. To solve it, this paper proposes a convolutional neural network with three sparse strategies named SCNN to classify DFC. Firstly, an element-wise filter is designed to impose sparse constraints on the DFC matrix by replacing the redundant elements with zeroes, where the DFC matrix is specially constructed to quantify the spatial and temporal variation of DFC. Secondly, a 1×1 convolutional filter is adopted to reduce the dimensionality of the sparse DFC matrix, and remove meaningless features resulted from zero elements in the subsequent convolution process. Finally, an extra sparse optimization classifier is employed to optimize the parameters of the above two filters, which can effectively improve the ability of SCNN to extract discriminative features. Experimental results on multiple resting-state fMRI datasets demonstrate that the proposed model provides a better classification performance of DFC compared with several state-of-the-art methods, and can identify the abnormal brain functional connectivity.

The brain functional connectivity classification based on deep learning is a research hotspot nowadays. However, the classification performance is far behind the demand of clinical applications. To alleviate the problem, this paper proposes a multiview deep learning method for brain functional connectivity classification. Firstly, the proposed method adopts multiple brain atlases to identify brain regions and thereby builds different brain functional connectivity of different views. Secondly, it uses a multiview feature selection strategy to select out the most discriminative features of each view with the assistance of other views. Then, it trains a stacked autoencoder to extract deep features of the brain functional connectivity of each view. At last, it utilizes a multiview fusion strategy to take full advantage of complementary information of different views for brain functional connectivity classification. The proposed method has been compared with several deep learning-based brain functional connectivity classification methods on three public datasets of neuropsychiatric disorders. The experimental results have validated the superior performance of the proposed method.

脑效应连接(Effective connectivity, EC)网络是人脑连接组研究中一项重要的研究课题,识别脑效应连接网络已成为评价正常脑功能及其与神经退化疾病相关损伤的一种有效手段.针对从功能性磁共振成像数据中进行脑效应连接网络的学习问题,提出了一种将多源信息与蚁群优化过程相融合的学习方法.新方法首先利用弥散张量成像数据获取感兴趣区域的结构约束信息,并利用正相关的皮尔森信息来压缩蚁群搜索的空间,以避免蚁群的许多不必要的搜索;然后在蚁群随机搜索中通过将体素联合激活信息融合于启发函数中,以增强蚂蚁搜索的目的性,改进算法的优化效率.实验结果验证了所提策略的有效性,与最新的同类算法相比,新算法在保持较快收敛速度的前提下,具有更好的求解质量.

为了有效应对脑功能连接高维小样本性给分类模型构建带来的挑战,得到与脑疾病诊断相关的重要特征,提出基于鱼群算法的脑功能连接邻域粗糙集特征归约方法.该方法建立脑功能连接数据的邻域决策表;依据特征的依赖度将鱼个体初始化为候选的脑功能连接特征子集,并采用综合特征子集依赖度和特征子集长度的适应度函数对鱼个体进行评价;在种群优化过程中,执行觅食、聚集、追尾机制,以及交叉和迁徙2个新机制来不断搜索最优的特征子集.在3种脑疾病功能磁共振脑成像(fMRI)数据集上,将所提方法与多种已有的特征归约方法进行对比实验.结果表明,该方法是有效的脑功能连接特征归约方法,可以有效降低脑功能连接数据的维度,获得分类判别能力较强的脑功能连接特征.

The diversity maintenance of nondominated solutions is crucial for solving multiobjective optimization problems. The grid strategy is a very effective way to maintain the diversity of nodominated solutions, but the existing grid strategies all adopt single-layer grid structure, which has weak ability for judging the distribution of nodominated solutions in the hyperboxes with the same crowding degree. To further explore the ability of the grid strategy for maintaining the diversity of nondominated solutions, this paper presents a new diversity maintenance strategy based on the double granularity grid. The double granularity grid strategy firstly partitions the hyperboxes with the same largest crowding degree into fine granularity hyperboxes. Then, it selects nondominated individual solutions according to the solution distribution in both coarse and fine granularity hyperboxes, which can avoids randomness for selecting individual solutions in the single grid structure. To validate the performance of the double granularity grid strategy, we first integrated it with two famous algorithms, then tested the two integration algorithms by comparing them with the original algorithms and four other state-of-the-art algorithms.The experimental results validate the powerful advantages of the proposed double granularity grid strategy.

针对群智能聚类方法在蛋白质相互作用网络功能模块检测问题上运行时间长的不足,本文提出了一种基于烟花算法的蛋白质相互作用网络功能模块检测方法(Fireworks Algorithm for Functional Module Detection in Protein-protein Interaction Net-works,简称FWA-FMD).首先结合蛋白质相互作用网络的拓扑结构信息和基因本体的功能注释信息,基于标签传播思想将每个烟花个体初始化为一种候选的功能模块划分.其次在每一代进化过程中,利用具有局部搜索和全局搜索自调整能力的爆炸操作对每个烟花个体进行优化,并同时采用精英保留和轮盘赌策略选择下一代烟花个体.最后通过将最优烟花个体中标签相同的节点划分到同一功能模块,以得到最终的功能模块检测结果.在酵母菌和人类两个物种的4个公共蛋白质相互作用网络数据集上的功能模块检测结果,分别用两种标准功能模块数据集作为基准来评价的实验表明:FWA-FMD算法不但求解时间少于遗传算法、蚁群算法和细菌觅食算法,而且在多项评价指标上与一些代表性算法相比都具有明显的优势,能够更好地识别功能模块.

Bacterial foraging optimization (BFO) has been proved to be an efficient optimization method and successfully applied to a variety of fields in the real world. In BFO, the chemotaxis process is a complex and close combination of swimming and tumbling and plays a crucial role in searching better solutions. A previous study has modeled the dynamics of the chemotaxis mechanism mathematically and investigated the stability and convergence behavior of the chemotaxis dynamics over the one-dimensional objective function by Lyapunov stability theorem. However, this study appears to be very limited from a practical point of view, and how to extend their study to the multi-dimensional objective function is a challenge. To solve it, we present a stability analysis of chemotaxis dynamics in BFO over the multi-dimensional objective function in this paper. First, the general mathematical model of the chemotaxis mechanism over the multi-dimensional objective function is created. Secondly, this paper uses the general descent search to analyze the general mathematical model and points out two necessary conditions for avoiding the bacterium to trap into a non-optimal solution. And then, the stability and convergence of the chemotaxis dynamics, represented by the general mathematical model, are proved by using Lyapunov stability theorem. Finally, empirical research is conducted to validate the above theoretical analysis.

When the dimension of text data is high, the regularized extreme learning machine ( ELM) of single hidden layer structure has not enough ability to express feature in the text classification. To solve the problem, this paper presented a text classification method based on multi-layer extreme learning machine ( ML-ELM) . First, the method used the compressed representation of extreme learning machine-based auto-encoder ( ELM-AE) to reduce the dimension of the text data. Then, the structure of the multi-hidden was used to represent high-level features in the text data, and the method of least squares was used to classify the text data. The experimental results on Reuters, 20newsgroup and Fudan University Chinese Corpus datasets show that this algorithm has a good classification performance compared with other algorithms.

The bat algorithm was used to detect the functional modules in protein-protein interaction networks (PPINs), in order to get better protein functional modules and reveal the function of proteins. The position of each bat individual represents a candidate functional module partition. Each protein node in PPIN and all its neighbor nodes form an ordered adjacency list and the population is initialized by random walk coding method in the ordered adjacency list. Four kinds of optimization mechanisms, namely directional local disturbance, random disturbance, adaptive variation based on distance and frequency, natural selection, are designed for the random optimization of solutions in the process of population optimization. The comparison experiments of the proposed algorithm and six classical algorithms were conducted on five yeast PPIN datasets having different scales. Results showed that many functional modules detected by the proposed method matched the standard modules and the evaluation indexes including coverage, recall, sensitivity, positive predictive value and accuracy were outstanding, which verified the validity of the proposed method.

脑效应连接网络学习是人脑连接组研究的一个重要研究课题,准确识别脑效应连接网络对于脑疾病的早期诊断以及病理研究具有重要意义.本文将萤火虫算法与贝叶斯网相结合,提出了一种带有繁殖机制的脑效应连接网络萤火虫学习方法.新方法使用K2评分作为目标函数来衡量萤火虫个体的绝对亮度,利用萤火虫种群的寻优来完成脑效应连接网络的学习,并利用繁殖机制对种群实施进一步的优化.首先将一种仅含少数边的脑效应连接网络表示成一个萤火虫个体,并通过萤火虫个体的定向移动操作以及随机移动操作逐步构建脑效应连接网络;然后每经过一定代数的寻优后,萤火虫种群执行一次繁殖过程,以优化效应连接网络的质量.最后,当算法收敛时,将萤火虫种群中绝对亮度最高个体所代表的网络结构作为学习到的最优脑效应连接网络.在多组模拟数据集上的实验结果验证了新算法中繁殖机制的有效性,且与其它算法相比,新算法具有明显优势.在真实数据上的实验也表明了算法的潜在实用性.

Identifying functional modules in PPI networks contributes greatly to the understanding of cellular functions and mechanisms. Recently, the swarm intelligence-based approaches have become effective ways for detecting functional modules in PPI networks. This paper presents a new computational approach based on bacterial foraging optimization for functional module detection in PPI networks (called BFO-FMD). In BFO-FMD, each bacterium represents a candidate module partition encoded as a directed graph, which is first initialized by a random-walk behavior according to the topological and functional information between protein nodes. Then, BFO-FMD utilizes four principal biological mechanisms, chemotaxis, conjugation, reproduction, and elimination and dispersal to search for better protein module partitions. To verify the performance of BFO-FMD, we compared it with several other typical methods on three common yeast datasets. The experimental results demonstrate the excellent performances of BFO-FMD in terms of various evaluation metrics. BFO-FMD achieves outstanding Recall, F-measure, and PPV while performing very well in terms of other metrics. Thus, it can accurately predict protein modules and help biologists to find some novel biological insights.

A Bayesian network (BN) is an important probabilistic model in the field of artificial intelligence and a powerful formalism used to describe uncertainty in the real world. As science and technology develop, considerable data on complex systems have been acquired by various means, which presents a significant challenge regarding how to accurately and robustly learn a network structure for a complex system. To address this challenge, many BN structure learning methods based on swarm intelligence have been developed. In this study, we perform a systematic comparison of three typical methods based on ant colony optimization, artificial bee colony algorithm, and bacterial foraging optimization. First, we analyze and summarize their main characteristics from the perspective of stochastic searching. Second, we conduct thorough experimental comparisons to examine the roles of different mechanisms in each method by means of multiaspect metrics, i.e., the K2 score, structural differences, and execution time. Next, we perform further experiments to validate the robustness of different algorithms on some benchmark data sets with noise. Finally, we present the prospects and references for researchers who are engaged in learning BN networks.