Scalable evolutionary clustering algorithm with Self Adaptive Genetic Operators

Clustering method, which groups thousands of genes by their similarities of expression levels, has been used for identifying unknown functions of genes. Fuzzy clustering method that is one category of clustering assigns one sample to multiple groups according to their membership degrees. It is more appropriate than hard clustering algorithms for analyzing gene expression profiles since single gene might involve multiple genetic functions. However, general clustering methods have problems that they are sensitive to initialization and can be trapped into local optima. To solve the problems, we propose an evolutionary fuzzy clustering algorithm with knowledge-based evaluation. It uses a genetic algorithm for clustering and prior knowledge of data for evaluation. Yeast cell-cycle dataset has been used for experiments to show the usefulness of the proposed method.KeywordsFuzzy ClusterMembership DegreeCluster PartitionAnalyze Gene Expression ProfileFuzzy Cluster MethodThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

A multi-objective adaptive evolutionary algorithm to extract communities in networks

Broad learning approach to Surrogate-Assisted Multi-Objective evolutionary fuzzy clustering algorithm based on reference points for color image segmentation

In this paper, a self-adaptive evolutionary clustering algorithm is presented. This algorithm uses the evolutionary programming (EP) to search the optimal clustering and bases on the principles of the K-means algorithm. The proposed self-adaptive evolutionary (SAEP) clustering algorithm self-adapts the vector of the step size appropriate for each parent. This is different from other genetic-based algorithms. The algorithm can minimize the degeneracy in the evolutionary process. The experimental results show that the KSAE clustering algorithm is efficient in the unsupervised classification of the multispectral remote sensing image.

In order to further enhance the performance of syntax parsing, for the shortcomings of hidden Markov model (HMM) in the parameter optimization, an improved syntax parsing method based on adaptive genetic annealing and HMM was presented. First, an adaptive hybrid genetic annealing algorithm was adopted to optimize HMM initial parameters. Second, the improved HMM was trained by Baum Welch algorithm, and then a modified Viterbi algorithm was used to recognize various types of phrases at the same layer, finally a hierarchical analysis algorithm and Viterbi algorithm were combined together to solve hierarchy and recursion in the sentence. In the adaptive genetic annealing HMM algorithm, genetic operators and parameters of simulated annealing (SA) were first respectively improved, subpopulations were classified according to the adaptive crossover and mutation probability of GA in order to realize the multi-group parallel search and information exchange, which could avoid premature and accelerate convergence, then SA was taken as a GA operator to strengthen the local search capability. Compared with several new approaches, Fβ = 1 value is averagely increased by 3%. The experiment results prove that this method is very effective for syntactic parsing.

Multi-robot coalition formation (MRCF) problem deals with the formation of subsets of robotic to handle a particular task. In such a system, every task is executed by multiple robots. Thus, cooperation and coordination among the robots is very important. One of the key issues to be investigated for smooth operation of a multi-robot systems is finding an optimal task allocation among the suitably formed robot groups (sub sets). Considering the complete execution of available tasks, the problem of assigning available resources (robot features) to the tasks is computationally complex, which may further increase as number of tasks increases. Genetic algorithms (GA) have been found quite efficient in solving such complex computational problems. There are several algorithms based on GA to solve MRCF problems but none of them have considered the dynamic variants. Thus we apply immigrants based GAs viz. RIGA (random immigrants genetic algorithm) and EIGA (elitism based immigrants genetic algorithm) to optimal task allocation in MRCF problem. Comparative performance evaluation has been made with respect to SGA (standard genetic algorithm). Finally, we report a novel use of these algorithms making them adaptive with certain modification in their traditional attributes by adaptively choosing the parameters of genetic operators. We name them as aRIGA (adaptive RIGA) and aEIGA (adaptive EIGA). Simulations experiments have demonstrated that RIGA and EIGA produces better solutions then SGA in both the cases (with fixed and adaptive genetic operators). Among them, EIGA and aEIGA outperforms RIGA and aRIGA respectively.

The number of objectives in many-objective optimization problems (MaOPs) is typically high and evolutionary algorithms face severe difficulties in solving such problems. In this paper, we propose a new scalable evolutionary algorithm, called evolutionary path control strategy (EPCS), for solving MaOPs. The central component of our algorithm is the use of a reference vector that helps simultaneously minimizing all the objectives of an MaOP. In doing so, EPCS employs a new fitness assignment strategy for survival selection. This strategy consists of two procedures and our algorithm applies them sequentially. It encourages a population of solutions to follow a certain path reaching toward the Pareto optimal front. The essence of our strategy is that it reduces the number of nondominated solutions to increase selection pressure in evolution. Furthermore, unlike previous work, EPCS is able to apply the classical Pareto-dominance relation with the new fitness assignment strategy. Our algorithm has been tested extensively on several scalable test problems, namely five DTLZ problems with 5 to 40 objectives and six WFG problems with 2 to 13 objectives. Furthermore, the algorithm has been tested on six CEC09 problems having 2 or 3 objectives. The experimental results show that EPCS is capable of finding better solutions compared to other existing algorithms for problems with an increasing number of objectives.

Evolutionary affinity propagation, an evolutionary clustering algorithm that groups data points by exchanging messages on a factor graph, is proposed. The algorithm promotes temporal smoothness of the clustering solutions at distinct temporal snapshots by linking variable nodes of the graph across time, and is capable of detecting cluster births and deaths. Unlike most existing evolutionary clustering methods that require additional processing in order to approximate the number of clusters, evolutionary affinity propagation determines the number of clusters automatically. A comparison with existing methods on simulated and experimental data demonstrates accuracy and robustness of the proposed framework.

Based on the assembly line scheduling problem, an improved adaptive genetic algorithm is proposed to solve the problem that small population genetic algorithm is easy to fall into local optimal solution. In the improved genetic algorithm, the mutation rate is increased in the early iteration to improve the diversity of offspring, and the mutation rate is reduced in the later iteration to retain effective genes. The improved roulette selection method is used to solve the problem that value of optimization objectives is large and single change of it is small. In order to improve the local search ability and computational speed of the algorithm, an adaptive genetic operator is used to dynamically adjust the crossover operator in the evolution process. The feasibility of the small population adaptive genetic algorithm is verified by experiments, and the performance is compared.

Associated evolution of a support vector machine-based classifier for pedestrian detection

Evolutionary multi-objective clustering (EMOC) algorithms have gained popularity recently, as they can obtain a set of clustering solutions in a single run by optimizing multiple objectives. Particularly, in one type of EMOC algorithm, the number of clusters k is taken as one of the multiple objectives to obtain a set of clustering solutions with different k. However, the numbers of clusters k and other objectives are not always in conflict, so it is impossible to obtain the clustering solutions with all different k in a single run. Therefore, evolutionary multi-objective k-clustering (EMO-KC) has recently been proposed to ensure this conflict. However, EMO-KC could not obtain good clustering accuracy on high-dimensional datasets. Moreover, EMO-KC’s validity is not ensured as one of its objectives (SSDexp, which is transformed from the sum of squared distances (SSD)) could not be effectively optimized and it could not avoid invalid solutions in its initialization. In this paper, an improved evolutionary multi-objective clustering algorithm based on autoencoder (AE-IEMOKC) is proposed to improve the accuracy and ensure the validity of EMO-KC. The proposed AE-IEMOKC is established by combining an autoencoder with an improved version of EMO-KC (IEMO-KC) for better accuracy, where IEMO-KC is improved based on EMO-KC by proposing a scaling factor to help effectively optimize the objective of SSDexp and introducing a valid initialization to avoid the invalid solutions. Experimental results on several datasets demonstrate the accuracy and validity of AE-IEMOKC. The results of this paper may provide some useful information for other EMOC algorithms to improve accuracy and convergence.

Due to the complex structure of observation based traffic accident data and the absence of an analytic model to define their characteristics, different models are required. Accident characteristics have been modeled for different road segments with two different methods: evolutionary data clustering method and resilient neural networks. In the first method, observation data was clustered using an evolutionary search-based clustering algorithm. The first method is based on determining whether observation based test data have the conditions of a possible death or injury accident based on the distance to the cluster centers obtained. The second one is a regression method that predicts whether an accident will cause death or injury according to observation based traffic data in test road segments by using resilient neural networks. Experiment results show that data analysis methods are very effective in determining the existence of the conditions that may cause accidents resulting in death or injury.

The search for efficient and predictive methods to describe the protein folding process at the all-atom level remains an important grand-computational challenge. The development of multi-teraflop architectures, such as the IBM BlueGene used in this study, has been motivated in part by the large computational requirements of such studies. Here we report the predictive all-atom folding of the forty-amino acid HIV accessory protein using an evolutionary stochastic optimization technique. We implemented the optimization method as a master-client model on an IBM BlueGene, where the algorithm scales near perfectly from 64 to 4096 processors in virtual processor mode. Starting from a completely extended conformation, we optimize a population of 64 conformations of the protein in our all-atom free-energy model PFF01. Using 2048 processors the algorithm predictively folds the protein to a near-native conformation with an RMS deviation of 3.43 A in < 24 h.

Every day, humans generate approximately 2.5 billion gigabytes of new data. While this data is crucial in applications such as drug development and cybersecurity, it is mostly unstructured and unlabeled which makes it difficult to process effectively. Businesses and governments alike solve this problem by using machine learning algorithms to characterize the data and make decisions with it. However, conventional machine learning processes consume excessive computational resources and time: one of the most prominent, K-means clustering, has an approximate time complexity of O(n^2), where n is the input data size. This quadratic time complexity causes scalability problems with large datasets. In this project, I have created an improved, more scalable clustering algorithm by leveraging quantum computation. The approach (new Q-means) involves an Angle Embedding Feature Map, an d-dimensional SWAP Test, and a dynamic, threshold distance-based termination criteria. After implementing this algorithm using the IBM Qiskit SDK, the new Q-means algorithm was run several times on different dimensional datasets. On an average, the new Q-means performed 18% better than the K-means algorithm based on the Adjusted Rand Index. In terms of the simulation execution time, the new Q-means was about six times faster than the existing Q-means due to a multi-threaded implementation and faster convergence. The time complexity of the new Q-means without a Quantum Random Access Memory (QRAM) is O(n^2). While the existing Q-means algorithm, which utilizes QRAM, has a time complexity of O(n(log n)^p), the time complexity of the new Q-means with QRAM is significantly better—O(n).

Finding clusters in data is a challenging problem especially when the clusters are being of widely varied shapes, sizes, and densities. Herein a new scalable clustering technique which addresses all these issues is proposed. In data mining, the purpose of data clustering is to identify useful patterns in the underlying dataset. Within the last several years, many clustering algorithms have been proposed in this area of research. Among all these proposed methods, density clustering methods are the most important due to their high ability to detect arbitrary shaped clusters. Moreover these methods often show good noise-handling capabilities, where clusters are defined as regions of typical densities separated by low or no density regions. In this paper, we aim at enhancing the well-known algorithm DBSCAN, to make it scalable and able to discover clusters from uneven datasets in which clusters are regions of homogenous densities. We achieved the scalability of the proposed algorithm by using the k-means algorithm to get initial partition of the dataset, applying the enhanced DBSCAN on each partition, and then using a merging process to get the actual natural number of clusters in the underlying dataset. This means the proposed algorithm consists of three stages. Experimental results using synthetic datasets show that the proposed clustering algorithm is faster and more scalable than the enhanced DBSCAN counterpart.