Results 1 
9 of
9
Clustering in the presence of bridgenodes
 Proc of SDM’06: SIAM Int’l Conf on Data Mining
, 2006
"... In this paper, we study the illeffects of bridgenodes, which causes many dissimilar objects to be placed together in the same cluster by existing clustering algorithms. We offer two new metrics for measuring how well a clustering algorithm handles the presence of bridgenodes. We also illustrate ho ..."
Abstract

Cited by 3 (1 self)
 Add to MetaCart
In this paper, we study the illeffects of bridgenodes, which causes many dissimilar objects to be placed together in the same cluster by existing clustering algorithms. We offer two new metrics for measuring how well a clustering algorithm handles the presence of bridgenodes. We also illustrate how algorithms that produce overlapping clusters help to alleviate the effect of bridgenodes and form more meaningful clusters. However, if there is too much overlap, the clusters become less informative. To address this problem, we present a novel clustering algorithm called MINCUT. Our experimental results with real data sets show that the MINCUT algorithm leads to purer clusters that have very little overlap. 1.
The Observable Part of a Network
"... Abstract — The union of all shortest path trees G∪spt is the maximally observable part of a network when traffic follows shortest paths. Overlay networks such as peer to peer networks or virtual private networks can be regarded as a subgraph of G∪spt. We investigate properties of G∪spt in different ..."
Abstract

Cited by 2 (2 self)
 Add to MetaCart
Abstract — The union of all shortest path trees G∪spt is the maximally observable part of a network when traffic follows shortest paths. Overlay networks such as peer to peer networks or virtual private networks can be regarded as a subgraph of G∪spt. We investigate properties of G∪spt in different underlying topologies with regular i.i.d. link weights. In particular, we show that the overlay G∪spt in an ErdösRényi random graph Gp (N) log N is a connected Gpc (N) where pc ∼ is the critical link N density, an observation with potential for adhoc networks. Shortest paths and, thus also the overlay G∪spt, can be controlled by link weights. By tuning the power exponent α of polynomial link weights in different underlying graphs, the phase transitions in the structure of G∪spt are shown by simulations to follow a same universal curve FT (α) =Pr[G∪spt is a tree]. The existence of a controllable phase transition in networks may allow network operators to steer and balance flows in their network. The structure of G∪spt in terms of the extreme value index α is further examined together with its spectrum, the eigenvalues of the corresponding adjacency matrix of G∪spt. Index Terms — Overlay, observability, union of shortest paths I.
Tree Augmented Classification of Binary Data Minimizing Stochastic Complexity
, 2002
"... We establish the algorithms and procedures that augment by trees the classfiers of binary feature vectors in (Gyllenberg et. al. 1993, 1997, Gyllenberg et. al. 1999 and Gyllenberg and Koski 2002). The notion of augmenting a classifier by a tree is due to (Chow and Liu 1968) and in a more extensive f ..."
Abstract

Cited by 1 (1 self)
 Add to MetaCart
We establish the algorithms and procedures that augment by trees the classfiers of binary feature vectors in (Gyllenberg et. al. 1993, 1997, Gyllenberg et. al. 1999 and Gyllenberg and Koski 2002). The notion of augmenting a classifier by a tree is due to (Chow and Liu 1968) and in a more extensive form due to (Friedman et. al. 1997). These techniques will in another report be primarily applied to unsupervised classification of bacterial DNA fingerprints (or electrophoretic patterns), c.f., (Gyllenberg and Koski 2001 (a), Rademaker et. al. 1999). By classification we mean here both the (unsupervised) procedures of finding the classes in (training) data of items as well as the actual outcome of the procedure, i.e., a partitioning of the items. By identification we mean the procedures for finding the assignment of items in classes, preestablished in one way or the other. The distinction should be clear, although the algorithms of classification as given in the sequel will also...
Clustering in the Presence of BridgeNodes
"... In this paper, we study the illeffects of bridgenodes, which causes many dissimilar objects to be placed together in the same cluster by existing clustering algorithms. We offer two new metrics for measuring how well a clustering algorithm handles the presence of bridgenodes. We also illustrate ho ..."
Abstract
 Add to MetaCart
In this paper, we study the illeffects of bridgenodes, which causes many dissimilar objects to be placed together in the same cluster by existing clustering algorithms. We offer two new metrics for measuring how well a clustering algorithm handles the presence of bridgenodes. We also illustrate how algorithms that produce overlapping clusters help to alleviate the effect of bridgenodes and form more meaningful clusters. However, if there is too much overlap, the clusters become less informative. To address this problem, we present a novel clustering algorithm called MINCUT. Our experimental results with real data sets show that the MINCUT algorithm leads to purer clusters that have very little overlap. 1.
Routing In An Awg Based Optical Packet Switch
, 2003
"... For the next generation of the optical internet, focus is now moving from circuit switched networks, which occupy a wavelength continuously regardless of the demand at that time, towards optical packet/burst switching. By only occupying a wavelength when data is to be transmitted, a more efficient u ..."
Abstract
 Add to MetaCart
For the next generation of the optical internet, focus is now moving from circuit switched networks, which occupy a wavelength continuously regardless of the demand at that time, towards optical packet/burst switching. By only occupying a wavelength when data is to be transmitted, a more efficient utilisation of bandwidth in optical fibres is strived for. As bandwidth in fibres keeps increasing, the bottleneck of the optical network is now moving towards the switching node, since evolution of electronic routers cannot follow the speed of bandwidth increase. Thus a key component in these novel networks is the optical node. Through this node we want to switch traffic very fast and reliable, preferably transparent. Lack of efficient and practically realisable optical memory however makes migration from electronic routers to optical routers a nonstraightforward transition. In most optical nodes payload traffic can be switched transparently, whilst control information (e.g. in a header, on a control channel) is still converted to the electronic domain in every node, since optical processing is far from mature. In this paper we present a possible architecture for such a node, combining Array Waveguide Gratings and alloptical tuneable wavelength converters. The concept of this switch is explained and the node is evaluated in terms of loss rate. We will see that an inherent problem of this switch is its internal blocking. This drawback can be greatly overcome by using an intelligent and efficient wavelength assignment algorithm within the node. Simulation of slotted operation will give some numerical results.
TargetOriented Routing Algorithm Based on Sequential Coordinates for Autonomous Wireless Sensor Network
"... Abstract—Wireless sensors implementation in process automation applications is a forwarding step for wireless sensor network. Autonomous network structure is considered as an option for such implementation. Autonomous wireless sensor/actuator networks require a targetoriented routing algorithm. In ..."
Abstract
 Add to MetaCart
Abstract—Wireless sensors implementation in process automation applications is a forwarding step for wireless sensor network. Autonomous network structure is considered as an option for such implementation. Autonomous wireless sensor/actuator networks require a targetoriented routing algorithm. In the first section the perception of autonomous network with an example is explained. It is clarified which features from the routing algorithm are expected. In the second section, Sequential Coordinate Routing Algorithm (SCAR) is proposed and its development, functionality and properties are discussed. Besides the targetoriented property as a main feature of the SCAR, based on the mathematical claim and its proof, it is shown how the minimum energy consumption is taken into consideration. By realizing the core of the routing algorithm, it is presented that not only void problem does not exist like other algorithms but also it is easy to compute. By this algorithm any two nodes are able to communicate to each other without need to pass message through central node of the network. Index Terms—Autonomous network, central network, wireless sensor network, targetoriented routing algorithm, sequential coordinate routing algorithm I.
ROUTING BALANCED COMMUNICATIONS ON HAMILTON DECOMPOSABLE NETWORKS
, 2003
"... Communicated by JeanClaude Bermond In 10  the authors proved upper bounds for the arccongestion and wavelength number of any permutation demand on a bidirected ring. In this note, we give generalizations of their results in two directions. The first one is that instead of considering only permut ..."
Abstract
 Add to MetaCart
Communicated by JeanClaude Bermond In 10  the authors proved upper bounds for the arccongestion and wavelength number of any permutation demand on a bidirected ring. In this note, we give generalizations of their results in two directions. The first one is that instead of considering only permutation demands we consider any balanced demand, and the second one is that instezul of the ring network we consider any Hamilton decomposable network. Thus, we obtain upper bounds (which are best possible in general) for the arccongestion and wavelength number of any balanced demand on a Hamilton decomposable network. As a special case, we obtain upper bounds on arc and edgeforwarding indices of Hamilton decomposable networks that are in many caises better than the known ones.
Clustering in the Presence of BridgeNodes
"... In this paper, we study the illeffects of bridgenodes, which causes many dissimilar objects to be placed together in the same cluster by existing clustering algorithms. We offer two new metrics for measuring how well a clustering algorithm handles the presence of bridgenodes. We also illustrate ho ..."
Abstract
 Add to MetaCart
In this paper, we study the illeffects of bridgenodes, which causes many dissimilar objects to be placed together in the same cluster by existing clustering algorithms. We offer two new metrics for measuring how well a clustering algorithm handles the presence of bridgenodes. We also illustrate how algorithms that produce overlapping clusters help to alleviate the effect of bridgenodes and form more meaningful clusters. However, if there is too much overlap, the clusters become less informative. To address this problem, we present a novel clustering algorithm called MINCUT. Our experimental results with real data sets show that the MINCUT algorithm leads to purer clusters that have very little overlap. 1.
1The Observable Part of a Network
"... Abstract—The union of all shortest path trees G∪spt is the maximally observable part of a network when traffic follows shortest paths. Overlay networks such as peer to peer networks or virtual private networks can be regarded as a subgraph of G∪spt. We investigate properties of G∪spt in different un ..."
Abstract
 Add to MetaCart
Abstract—The union of all shortest path trees G∪spt is the maximally observable part of a network when traffic follows shortest paths. Overlay networks such as peer to peer networks or virtual private networks can be regarded as a subgraph of G∪spt. We investigate properties of G∪spt in different underlying topologies with regular i.i.d. link weights. In particular, we show that the overlay G∪spt in an ErdösRényi random graph Gp (N) is a connected Gpc (N) where pc ∼ logNN is the critical link density, an observation with potential for adhoc networks. Shortest paths and, thus also the overlay G∪spt, can be controlled by link weights. By tuning the power exponent α of polynomial link weights in different underlying graphs, the phase transitions in the structure of G∪spt are shown by simulations to follow a same universal curve FT (α) = Pr[G∪spt is a tree]. The existence of a controllable phase transition in networks may allow network operators to steer and balance flows in their network. The structure of G∪spt in terms of the extreme value index α is further examined together with its spectrum, the eigenvalues of the corresponding adjacency matrix of G∪spt. Index Terms—Overlay, observability, union of shortest paths I.