Clustering plays an important role in data mining. It helps to reveal intrinsic structure in data sets with little or no prior
knowledge. The approaches of clustering have received great attention in recent years. However many published
algorithms fail to do well in determining the number of cluster, finding arbitrary shapes of clusters or identifying the
presence of noise. In this paper we present an efficient clustering algorithm which employs the theory of grid, density
and fractal that can partition points in the same cluster with minimum change of fractal dimension meanwhile
maximizing the self-similarity in the clusters. We show via experiments that FDC can quickly deal with multidimensional
large data sets, identify the number of clusters, be capable of recognizing clusters of arbitrary shape and
furthermore explore some qualitative information from data sets.
Peer-to-Peer (P2P) systems are currently receiving considerable interest. However, as experience with P2P networks shows, the selfish behaviors of peers may lead to serious problems of P2P network, such as free-riding and white-washing. In order to solve these problems, there are increasing considerations on reputation system
design in the study of P2P networks. Most of the existing works is concerning probabilistic estimation or social
networks to evaluate the trustworthiness for a peer to others. However, these models can not be efficient all the
time. In this paper, our aim is to provide a general mechanism that can maximize P2P networks social welfare in
a way of Vickrey-Clarke-Groves family, while assuming every peer in P2P networks is rational and selfish, which
means they only concern about their own outcome. This mechanism has some desirable properties using an <i>O</i>(<i>n</i>)
algorithm: (1) incentive compatibility, every peer truly report its connection type; (2) individually rationality;
and (3) fully decentralized, we design a multiple-principal multiple-agent model, concerning about the service
provider and service requester individually.
Clustering or grouping of similar objects is one of the most widely used procedures in data mining, which has received
enormous attentions and many methods have been proposed in these recent decades. However these traditional clustering
algorithms require all the data objects to be located at one single site where it is analyzed. And such limitation cannot
face the challenge as nowadays monstrous sizes of data sets are often stored on different independently working
computers connected to each other via local or wide area networks instead of one single site. Therefore in this paper, we
propose a fully distributed clustering algorithm, called a fully distributed clustering based on fractal dimension
(FDCFD), which enables each site to collaborate in forming a global clustering model with low communication cost. The
main idea behind FDCFD is via calculating fractal dimension to group points in a cluster in such a way that none of the
points in the cluster changes the cluster's fractal dimension radically. In our theoretical analysis, we will demonstrate
that our approach can work very well for clustering data that is inherently distributed, collect information spread over
several local sites to form a global clustering meanwhile without communication costs and delays for transmitting.
An important goal in P2P networks is that all peers provide resources. However, free riding and tragedy of common are real issues in P2P networks. To resolve these problems, most of the existing work is concerning probabilistic estimation to evaluate the trustworthiness or mechanism design to provide incentive. Instead of design a protocol to solve free riding, we build a micro-payment architecture for these existing protocols using virtual currency which can be more precisely measured and easily be replaced by reputation or other tokens. Our system can avoid from long-term trust learning interactions and high cost of collecting and analyzing reputation information. It can also provide peers incentive to truly report their connection type and security to malicious attacks.