Zhang Yan-ping

Cross Covering Algorithm Based on Simulated Annealing

Structural machine learning method—covering algorithm (CA) possesses faster speed, lower complexity and higher precision. But construction of the weight of the neurons for new center of sphere domain is usually given a manmade criteria, could not follow the distribution of samples to achieve the optimal solution. In this paper, a new constructive algorithm which combines the cross covering algorithm and Simulated Annealing is presented. It gets the covering center according to the search of the simulated annealing theory. The results show that the algorithm can reduce the number of coverings with higher degree of recognition accuracy.

An Improved Target Immunization Strategy

Strategies for network immunization have been drawing wide interest in the area of complex networks during the past decade. The target strategy, in which degree values are used to determine vaccinating priority, has been supposed to be the most effective strategy. In many networks where large numbers of nodes exists with same degree values, the target strategy will face a problem of deciding the priorities of nodes with equal degree. This paper discussed a refined method to evaluate priority using risk-value. With risk-values, an improved target immunization strategy is proposed. Simulations under SIS model using real network dataset have proved a better efficiency for the improved strategy.

A New Algorithm of Detecting Communities by Topological Centrality

Community structure is an important common property of complex network which has much theoretical significance for analyzing structure, mastering function, detecting implicit scheme and predicting activities. This paper reviews some popular detecting community methods. NTCDC Algorithm is proposed by analyzing weigh of nodes’ and edges’ TC values. The algorithm not only improves the precision, but also faster than GN and fast algorithm proposed by Newman. This paper demonstrate that the algorithm is higher effective and faster at discovering community structure in both computer-generated network and real-world network data than GN and fast algorithm.

The Least Covering Algorithm

Learning system in machine learning is conducted to confirm the description of specific concept, according to a set of samples and background knowledge that teachers offer. In the point of epistemology , when study the samples, we always focus on the sample set, so nothing can be fabricated, in other words, if we only have a few samples, we can get limited knowledge after learning them, then it is impossible to distinguish every unknown situation. To get the principle which is close to the sample as much as possible, this paper puts forward the least covering principle of machine learning, which is the aim of the covering algorithm of multi-layered feedforward neural network; it also makes a study of the properties of least covering, then brings forward a geometry algorithm to get the least covering that is based on this; at last it gives the solving process of least covering using the programming method.

Quotient Space Theory about An Irreducible Iterated Function System

In this paper, we use the relations of quotient space theory and martingale theory to research the iterated function system that is fractal geometry images, and propose these conclusions: Given an irreducible iterated function system {X, w_i, p_ij; i, j = 1, 2, ..., n} , then exists a corresponding chain of quotient space {W_k = (X_k, mu_k, F_k); k = 1, 2, ...} and a martingale {(mu_k, F_k); k = 1, 2, ...} on the chain, therefore there are: 1) Assume P_k is a invariant subsets of W_k, P is a invariant subsets of W, then exists lim_krarrinfin P_k = P and the convergence is according to Hausdorff distance. 2) Assume mu_k is a invariant measure of F_k, mu is a invariant measure of F, then exists lim_krarrinfin mu_k = mu. 3) P_k is a support set of mu_k, P is a support set of mu. Namely we present the quotient approximation theorem about fractal geometry images, and build relations among chain of quotient space, martingale , fractal geometry images and Markovian process.