As a fundamental mathematical problem in the field of machine learning, the linear separability test still lacks a theoretically complete and computationally efficient method. This paper proposes and proves a sufficient and necessary condition for linear separability test based on a sphere model. The advantage of this test method is two-fold: (1) it provides not only a qualitative test of linear separability but also a quantitative analysis of the separability of linear separable instances; (2) it has low time cost and is more efficient than existing test methods. The proposed method is validated through a large number of experiments on benchmark datasets and artificial datasets, demonstrating both its correctness and efficiency.

The linearly separable problem is a fundamental problem in pattern classification. Firstly, from the perspective of spatial distribution, this paper focuses on the linear separability of a region dataset at the distribution level instead of the linearly separable issue between two datasets at the traditional category level. Firstly, the former can reflect the spatial distribution of real data, which is more helpful to its application in pattern classification. Secondly, based on spatial geometric theory, an adaptive construction method for testing the linear separability of a region dataset is demonstrated and designed. Finally, the corresponding computer algorithm is designed, and some simulation verification experiments are carried out based on some manual datasets and benchmark datasets. Experimental results show the correctness and effectiveness of the proposed method.

The purpose of supervised learning with temporal encoding for spiking neurons is to make the neurons emit a specific spike train encoded by precise firing times of spikes. The gradient-descent-based (GDB) learning methods are widely used and verified in the current research. Although the existing GDB multi-spike learning (or spike sequence learning) methods have good performance, they work in an offline manner and still have some limitations. This paper proposes an online GDB spike sequence learning method for spiking neurons that is based on the online adjustment mechanism of real biological neuron synapses. The method constructs error function and calculates the adjustment of synaptic weights as soon as the neurons emit a spike during their running process. We analyze and synthesize desired and actual output spikes to select appropriate input spikes in the calculation of weight adjustment in this paper. The experimental results show that our method obviously improves learning performance compared with the offline learning manner and has certain advantage on learning accuracy compared with other learning methods. Stronger learning ability determines that the method has large pattern storage capacity.

Aiming at the problems that the purpose of the meteorological observation data acquisition is weak,the redundancy of data is high,and the number of single values in the observation data interval is large,the precision of equivalence partitioning is low,an attribute reduction algorithm for Meteorological Observation data Interval-value based on Genetic Algorithm (MOIvGA) was proposed.Firstly,by improving the similarity degree of interval value,the proposed algorithm could be suitable for both single value equivalence relation judgment and interval value similarity analysis.Secondly,the convergence of the algorithm was improved by the improved adaptive genetic algorithm.Finally,the simulation experiments show that the number of the iterations of the proposed algorithm is reduced by 22,compared with the method which operated AGAv (Adaptive Genetic Attribute reduction) algorithm to solve the optimal value.In the time interval of 1 hour precipitation classification,the average classification accuracy of the MOIvGA ()t-Reduction in Interval-valued decision table based on Dependence) algorithm is 6.3％ higher than that of RIvD algorithm;the accuracy of no rain forecasting is increased by 7.13％;at the same time,the classification accuracy can be significantly impoved by the attribute subset received by operating the MOIvGA algorithm.Therefore,the MOIvGA algorithm can increase the convergence rate and the classification accuracy in the analysis of interval value meteorological observation data.

Nowadays, the big success of deep learning makes artificial neural network becoming a hot topic once again, and the size of neural networks’ structure is a key visual cue for structured learning. The greater network may get the study task done well, while it may increase network computation overhead easier and cost more. Hence, network construction is an important issue, as well as a difficult problem. In this paper, we proposed a novel sensitivity-based adaptive architecture pruning algorithm for Madalines. The algorithm establishes a pruning measure based on the network sensitivity to its structure variation and a minimal disturbance principle. The measure can be used to evaluate the performance loss due to its structure changes more or less. And the loss can be compensated by relearning. Thus, the new adaptive pruning mechanism is developed with measuring, pruning, and compensating. The simulation experimental results based on some benchmark data demonstrate that the pruning measure is rationality and the new algorithm is effective.

Focusing on the lack of efficient and practical learning algorithm for Binary Feedforward Neural Networks ( BFNN) , a novel learning algorithm by fusing the self?adaptations of both architecture and weight for training BFNN is proposed. Based on improving the methodology of Extreme Learning Machines ( ELM) , the algorithm can effectively train BFNNs with single hidden layer for solving classification problems. In order to satisfy training accuracy, the algorithm can automatically increase hidden neurons and adjust the neuron ’ s weights with the Perceptron Learning Rule. As to improve generalization accuracy, the algorithm can automatically, by establishing binary neuron’ s sensitivity as a tool for measuring the relevance of each hidden neuron, prune the least relevant hidden neuron with some compensation for information losing due to the pruning. Experiment results verified the feasibility and effectiveness of the proposed algorithm.

In this paper, we proposed a new sensitivity-based adaptive architecture pruning algorithm for Madalines. The algorithm establishes a pruning measure based on the network sensitivity to its structure variation and a minimal disturbance principle. The measure can be used to evaluate the performance loss due to its structure changes more or less. And the loss can be compensated by relearning. Thus, the new adaptive pruning mechanism is developed with measuring, pruning, and compensating.

In order to improve the learning mechanism of BFNNs, the paper firstly analyzes the failure mode of BFNNs trained by SBALR, which takes the form of a local cycle. And then by mean of the sensitivity theory, a disturbance learning algorithm is developed to make the BFNNs that suffering from learning failure to escape the local cycle. The new algorithm aims to keep the existing learning performance as much as possible. Experimental results demonstrate the effectiveness of the new algorithm on both learning effect and learning efficiency.

There are many evolutionary algorithms for numerical optimization problems. However, the universality and robustness of existing algorithms are still unsatisfactory, and the performance of these algorithms deteriorates significantly as the dimensionality of the optimization problems increases. In this paper, an ensemble of evolution algorithm based on self-adaptive learning population search techniques (EEA-SLPS) is presented to overcome these defects on the numerical optimization problems. The EEA-SLPS integrates three self-adaptive learning based stochastic search algorithms which are termed as sub-algorithms. In the EEA-SLPS, the population is divided into three sub-populations, and the sub-algorithms are employed to evolve the sub-populations in parallel. Among the three sub-algorithms, one is designed in this paper and the other two are proposed by relevant literature, and eighteen information exchanging manners (IEMs) between sub-populations are investigated in order to make use of the sub-algorithms efficiently. We have found the most suitable IEM for the EEA-SLPS according to experimental investigations. Finally, the EEA-SLPS is tested on a suite of 26 bound-constrained functions with low and high dimensionality. The experimental results indicate that the universality and robustness performance of EEA-SLPS is better. Meanwhile, the results clearly verify the advantages of EEA-SLPS on the numerical optimization problems with low or high dimensionality.

This paper presents a novel and effective approach for establishing a quantified output sensitivity of Binary Feedforward Neural Networks to weight and input perturbations. Firstly, analytical formulae are derived for computing a neuron׳s sensitivity by means of matrix and probability theories. Then, based on the neuron׳s sensitivity and the network׳s architecture feature, a bottom-up strategy is followed to compute the entire network׳s sensitivity. The proposed approach has the obvious advantages of higher generality, lower computational complexity, and yet much higher accuracy. Experimental results verify the correctness and effectiveness of the approach.

With the deep application of RFID, the deployment of the RFID system is gradually to have the large-scale, networked, distributed development trend.The previous abnormal data analysis and processing way can not meet with the need of data processing.According to the survey of international famous consulting company (Gartner) for the "information overload", results show that more than 90% of the enterprises think the enterprise competitiveness is closely related to its information capacity.In order to obtain the useful information, we must analyze, process the data, and exclude "exception" data.So, it is very important to realize the detection of abnormal data.This paper presents the RFID supply an abnormal data analysis and processing system of RFID supply chain based on distance, rules and middleware technology .This system is divided into three levels: data processing layer, anomaly analysis lay and graphical display layer.Finally, through the performance analysis of the system, the effectiveness of the system has been verified.

Many consumer products in the home environment are managed by the Wireless Sensor Networks (WSNs). However, the energy hole problem in the WSNs which with logical ring topology and uniformly distributed sensors is usually caused by the energy exhaustion of the sensors which distributed in the first radius range of the sink. This paper firstly analyzed the energy consumption model of the sensor, the data transmission model of the sensor, and the energy consumption distribution model of the WSNs. Then, a WSN Energy Hole Alleviating (WSNEHA) algorithm, which is based on the data forwarding and router selection strategy, is proposed. The WSNEHPA adopts the data forwarding and routing selection strategy to balance the energy consumption of the sensors in the first radius range of the sink. Experimental results demonstrate that WSNEHPA can efficiently balance the energy consumption of the sensors in the first radius range of the sink, and that the lifetime of the WSNs can be extended efficiently.

This paper proposes a new adaptive learning algorithm for Madalines based on a sensitivity measure that is established to investigate the effect of a Madaline weight adaptation on its output. The algorithm, following the basic idea of minimal disturbance as the MRII did, introduces an adaptation selection rule by means of the sensitivity measure to more accurately locate the weights in real need of adaptation. Experimental results on some benchmark data demonstrate that the proposed algorithm has much better learning performance than the MRII and the BP algorithms.

In recent years, the nearest neighbor search (NNS) problem has been widely used in various interesting applications. Locality-sensitive hashing (LSH), a popular algorithm for the approximate nearest neighbor problem, is proved to be an efficient method to solve the NNS problem in the high-dimensional and large-scale databases. Based on the scheme of p-stable LSH, this paper introduces a novel improvement algorithm called randomness-based locality-sensitive hashing (RLSH) based on p-stable LSH. Our proposed algorithm modifies the query strategy that it randomly selects a certain hash table to project the query point instead of mapping the query point into all hash tables in the period of the nearest neighbor query and reconstructs the candidate points for finding the nearest neighbors. This improvement strategy ensures that RLSH spends less time searching for the nearest neighbors than the p-stable LSH algorithm to keep a high recall. Besides, this strategy is proved to promote the diversity of the candidate points even with fewer hash tables. Experiments are executed on the synthetic dataset and open dataset. The results show that our method can cost less time consumption and less space requirements than the p-stable LSH while balancing the same recall.

The invention discloses a quantum-behaved particle swarm optimization (QPSO) based multi-objective dynamic workflow scheduling method, and belongs to the technical field of cloud computing. The method includes the steps: inputting a workflow and a QoS (quality of service) request; acquiring state information of virtual machines and transmission information among the virtual machines; setting a to-be-executed task set V', and setting objective functions of time, cost and reliability for a task schedule in the V'; allocating optimal resources to the to-be-executed tasks by the aid of QPSO, and judging whether total time, total cost and total reliability of task execution meet the QoS request of a user or not after the tasks are executed; dynamically updating the V', transmission speed among the virtual machines and operating speeds of the virtual machines. By means of dynamically partitioning the workflow and dynamically updating network bandwidth information, the optimal resources are allocated to the workflow tasks accurately, errors between the calculated time and actual execution time and the calculated cost and actual execution cost are reduced, time can be shortened, and cost is reduced while reliability is enhanced.

The purpose of supervised learning with temporal encoding for spiking neurons is to make the neurons emit a specific spike train encoded by the precise firing times of spikes. If only running time is considered, the supervised learning for a spiking neuron is equivalent to distinguishing the times of desired output spikes and the other time during the running process of the neuron through adjusting synaptic weights, which can be regarded as a classification problem. Based on this idea, this letter proposes a new supervised learning method for spiking neurons with temporal encoding; it first transforms the supervised learning into a classification problem and then solves the problem by using the perceptron learning rule. The experiment results show that the proposed method has higher learning accuracy and efficiency over the existing learning methods, so it is more powerful for solving complex and real-time problems.

An improved fuzzy c-means algorithm is put forward and applied to deal with meteorological data on top of the traditional fuzzy c-means algorithm.The proposed algorithm improves the classical fuzzy c-means algorithm (FCM) by adopting a novel strategy for selecting the initial cluster centers, to solve the problem that the traditional fuzzy c-means (FCM) clustering algorithm has difficulty in selecting the initial cluster centers.Furthermore, this paper introduces the features and the mining process of the open source data mining platform WEKA, while it doesn't implement the FCM algorithm.Considering this shortcoming of WEKA, we successfully implement the FCM algorithm and the advanced FCM algorithm taking advantage of the basic classes in WEKA.Finally, the experimental clustering results of meteorological data are given, which can exactly prove that our proposed algorithm will generate better clustering results than those of the K-Means algorithm and the traditional FCM algorithm.

The sensitivity of a neural network's output to its inputs' perturbations is an important measure for evaluating the network's performance. To make the sensitivity be a practical tool for designing and implementing Multilayer Perceptrons (MLPs), this paper proposes a general approach to quantify the sensitivity of MLPs. The sensitivity is defined as the mathematical expectation of absolute output deviations due to input perturbations with respect to all possible inputs, and computed following a bottom-up way, in which the sensitivity of a neuron is first considered and then is that of the entire network. The main contribution of the approach is that it requests a weak assumption on the input, that is, input elements need only to be independent of each other without being restricted to have a certain type of distribution and thus is more applicable to real applications. Some experimental results on artificial datasets and real datasets demonstrate the proposed approach is highly accurate.

This paper, with an aim at improving neural networks' generalization performance, proposes an effective neural network ensemble approach with two novel ideas. One is to apply neural networks' output sensitivity as a measure to evaluate neural networks' output diversity at the inputs near training samples so as to be able to select diverse individuals from a pool of well-trained neural networks; the other is to employ a learning mechanism to assign complementary weights for the combination of the selected individuals. Experimental results show that the proposed approach could construct a neural network ensemble with better generalization performance than that of each individual in the ensemble combining with all the other individuals, and than that of the ensembles with simply averaged weights.

This paper proposes a set of adaptive learning rules for binary feedforward neural networks (BFNNs) by means of the sensitivity measure that is established to investigate the effect of a BFNN's weight variation on its output. The rules are based on three basic adaptive learning principles: the benefit principle, the minimal disturbance principle, and the burden-sharing principle. In order to follow the benefit principle and the minimal disturbance principle, a neuron selection rule and a weight adaptation rule are developed. Besides, a learning control rule is developed to follow the burden-sharing principle. The advantage of the rules is that they can effectively guide the BFNN's learning to conduct constructive adaptations and avoid destructive ones. With these rules, a sensitivity-based adaptive learning (SBALR) algorithm for BFNNs is presented. Experimental results on a number of benchmark data demonstrate that the SBALR algorithm has better learning performance than the Madaline rule II and backpropagation algorithms.