Convolutional Neural Networks (CNNs) have shown remarkable performance in general object recognition tasks. In this paper, we propose a new model called EnsNet which is composed of one base CNN and multiple Fully Connected SubNetworks (FCSNs). model, the set feature-maps generated by last convolutional layer divided along channels into disjoint subsets, these subsets are assigned to FCSNs. Each FCSNs trained independent others so that it can predict class label from subset it. The output...

This paper gives a new sufficient condition for cellular neural networks with delay (DCNNs) to be completely stable. A fixed-point theorem and convergence of the Gauss-Seidel method play important roles in proof, while most conventional stability criteria were obtained by constructing Lyapunov functionals.

The second smallest eigenvalue of the Laplacian matrix, also known as algebraic connectivity, characterizes performance some dynamic processes on networks, such consensus in multiagent synchronization coupled oscillators, random walks graphs, and so on. In a network, for example, larger connectivity graph representing interactions between agents is, faster convergence speed representative algorithm is. This paper tackles problem finding graphs that maximize or locally space with fixed number...

Moderate water stress tomato cultivated hydroponically in the greenhouse contains high lycopene and very sensitive to storage temperatures. This study aimed observe effect of temperatures on content color quality parameters (both moderate no tomato). The increased with higher than 10 °C while relatively stable or slightly. lightness (L*) value decreased during 10, 15, 25 30 redness (a*), yellowness (b*), a*/b*, hue (h), chroma (C*) remained after 4 days those Storage above 15 both tomato....

This paper gives a new sufficient condition for complete stability of nonsymmetric cellular neural network (CNN). The convergence theorem the Gauss-Seidel method, which is an iterative technique solving linear algebraic equation, plays important role in our proof. It also shown that existence stable equilibrium point does not imply CNN.

Global convergence of the sequential minimal optimization (SMO) algorithm for support vector regression (SVR) is studied in this paper. Given l training samples, SVR formulated as a convex quadratic programming (QP) problem with pairs variables. We prove that if two variables violating optimality condition are chosen update each step and subproblems solved certain way, then SMO always stops within finite number iterations after finding an optimal solution. Also, efficient implementation...

This paper presents an 6-axis optical force sensor which can be used in fMRI. Recently, fMRIs are widely for studying human brain function. Simultaneous measurement of activity and peripheral information, such as grip force, enables more precise investigations studies motor However, conventional sensors cannot fMRI environment, since metal elements generate noise severely contaminate the signals An 2-axis has been developed using photo fibers by Tada et. al.(2002), that resolved these...

Decomposition methods are well-known techniques for solving quadratic programming (QP) problems arising in support vector machines (SVMs). In each iteration of a decomposition method, small number variables selected and QP problem with only the is solved. Since large matrix computations not required, applicable to problems. this paper, we will make rigorous analysis global convergence general SVMs. We first introduce relaxed version optimality condition then prove that method reaches...

Hierarchical alternating least squares (HALS) algorithms are efficient computational methods for nonnegative matrix factorization (NMF). Given an initial solution, HALS update the solution block by iteratively so that error decreases monotonically. However, rules in not well-defined. In addition, due to this problem, convergence of sequence solutions a stationary point cannot be proved theoretically. paper, we consider algorithm Frobenius norm-based NMF, and prove modified version has global...

This letter gives a new sufficient condition for nonsymmetric CNNs to have at least one stable equilibrium point. Existence of point is important because it necessary complete stability. It shown that our generalization previous result concerning the existence point, and can easily be applied space invariant with 3/spl times/3 neighborhood.

A novel method to simplify decision functions of support vector machines (SVMs) is proposed in this paper. In our method, a function determined first usual way by using all training samples. Next those vectors which contribute less the are excluded from Finally new obtained remaining Experimental results show that can effectively SVMs without reducing generalization capability.

This paper proposes a novel on-demand multipath routing protocol (AODVM-PSP) for mobile ad hoc network. We devised new utilization method with probability selection, which each node deliberately selects one of the routes to utilize. examined our under various network loads. evaluated Omnet simulator and performance is compared conventional methods. When load high, number control packets prodigious. The simulation results show that proposed has significant improvement over others in high load.

A novel algorithm for multi-class support vector machines (SVMs) is proposed in this paper. The tree constructed our consists of a series two-class SVMs. Considering both separability and balance, each iteration patterns are divided into two sets according to the distances between pairwise classes number class. This can well treat with unequally distributed problems. efficiency method verified by experimental results.

A novel method for training support vector machines (SVMs) is proposed to speed up the SVMs in test phase. It has three main steps. First, an SVM trained on all samples, thereby producing a number of vectors. Second, vectors, which contribute less shape decision surface, are excluded from set. Finally, re-trained only remaining samples. Compared initially SVM, efficiency finally highly improved, without system degradation.

We study global dynamical behavior of cellular neural networks (CNNs) consisting two cells. Since the output characteristic each cell is expressed by a piecewise-linear function, CNN with cells considered as planar system. present necessary and sufficient condition for such to be completely stable under assumptions that: 1) self-coupling coefficients take same value greater than one 2) biases are set zero. The explicitly in terms coupling between

Data grid consists of scattered computing and storage resources located dispersedly in the network. These large sized data sets are replicated more than one site for better availability to other nodes a grid. Downloading dataset from these locations have practical difficulties we find interest co-allocated download framework, which enables parallel multiple servers. In this paper, proposed dynamic co-allocation scheme transfer environment, copes up with highly inconsistent network server...

Nonnegative Matrix Factorization (NMF) with sparseness and smoothness constraints has attracted increasing attention. When these properties are considered, NMF is usually formulated as an optimization problem in which a linear combination of approximation error term some regularization terms must be minimized under the constraint that factor matrices nonnegative. In this paper, we focus our attention on measure based Euclidean distance propose new iterative method for solving those problems....