Satu Elisa Schaeffer

On the NP-Completeness of some graph cluster measures

Graph clustering is the problem of identifying sparsely connected dense subgraphs (clusters) in a given graph. Identifying clusters can be achieved by optimizing a fitness function that measures the quality of a cluster within the graph. Examples of such cluster measures include the conductance, the local and relative densities, and single cluster editing. We prove that the decision problems associated with the optimization tasks of finding clusters that are optimal with respect to these fitness measures are NP-complete.

Stochastic local clustering for massive graphs

Most graph-theoretical clustering algorithms require the complete adjacency relation of the graph representing the examined data. This is infeasible for very large graphs currently emerging in many application areas. We propose a local approach that computes clusters in graphs, one at a time, relying only on the neighborhoods of the vertices included in the current cluster candidate. This enables implementing a local and parameter-free algorithm. Approximate clusters may be identified quickly by heuristic methods. We report experimental results on clustering graphs using simulated annealing.

Local clustering of large graphs by approximate fiedler vectors

We address the problem of determining the natural neighbourhood of a given node i in a large nonunifom network G in a way that uses only local computations, i.e. without recourse to the full adjacency matrix of G. We view the problem as that of computing potential values in a diffusive system, where node i is fixed at zero potential, and the potentials at the other nodes are then induced by the adjacency relation of G. This point of view leads to a constrained spectral clustering approach. We observe that a gradient method for computing the respective Fiedler vector values at each node can be implemented in a local manner, leading to our eventual algorithm. The algorithm is evaluated experimentally using three types of nonuniform networks: randomised “caveman graphs”, a scientific collaboration network, and a small social interaction network.

Metrics and Models for Social Networks

Social networks can be modeled and analyzed in terms of graph theory. This chapter provides an overview of the mathematical modeling of social networks with an overview of the metrics used to characterize them and the models used to artificially mimic the formation of such networks. We discuss metrics based on distances, degrees, and neighborhoods as well as the use of such metrics to detect change in the network structure. We also discuss the kind of structural differences that distinguish social networks from other types of natural networks together with the implications of these differences about the way in which these networks function.

Dynamic Local Clustering for Hierarchical Ad Hoc Networks

Hierarchical, cluster-based routing greatly reduces routing table sizes compared to host-based routing, while reducing path efficiency by at most a constant factor. More importantly, the amount of routing related signalling traffic is reduced. On the other hand, address changes caused by nodes changing their cluster produces address management traffic. In this paper, we present a new local clustering method that produces dense and stable clusters, thereby minimizing address changes and allowing better and more stable network conditions for ad hoc routing

NAS Algorithm for Semantic Query Routing Systems in Complex Networks

The modern distributed systems are acquiring a great importance in our daily lives. Each day more transactions are conducted through devices which perform queries that are dependent on the reliability, availability and security of distributed applications. In addition, those systems show great dynamism as a result of the extremely complex and unpredictable interactions between the distributed components, making it practically impossible to evaluate their behavior. In this paper, we evaluate the performance of the NAS algorithm (Neighboring-Ant Search), which is an algorithm for distributed textual query routing based on the Ant Colony System metaheuristic and SemAnt algorithm, improved with a local topological characterization metric and a classic local exploration method called lookahead, with the aim of improving the performance of the distributed search. Our results show that including local information like the topological metric and the exploration method in the Neighboring-Ant Search algorithm improves its performance 40%, in terms of the number of hops needed to locate a set of resources in a scale-free network.

Load balancing by distributed optimisation in ad hoc networks

We approach the problem of load balancing for wireless multi-hop networks by distributed optimisation. We implement an approximation algorithm for minimising the maximum network congestion as a modification to the DSR routing protocol. The algorithm is based on shortest-path computations that are integrated into the DSR route discovery and maintenance process. The resulting Balanced Multipath Source Routing (BMSR) protocol does not need to disseminate global information throughout the network. Our simulations with the ns2 simulator show a gain of 14% to 69% in the throughput, depending on the setup, compared to DSR for a high network load.

Estimating The Size Of Peer-To-Peer Networks Using Lambert's W Function

SUMMARY In this work, we address the problem of locally estimating the size of a Peer-to-Peer (P2P) network using local information. We present a novel approach for estimating the size of a peer-to-peer (P2P) network, fitting the sum of new neighbors discovered at each iteration of a breadth-first search (BFS) with a logarithmic function, and then using Lambert’s W function to solve a root of a ln(n) + b − n = 0, where n is the network size. With rather little computation, we reach an estimation error of at most 10 percent, only allowing the BFS to iterate to the third level.

Local bilateral clustering for identifying research topics and groups from bibliographical data

The structure of scientific collaboration networks provides insight on the relationships between people and disciplines. In this paper, we study a bipartite graph connecting authors to publications and extract from it clusters of authors and articles, interpreting the author clusters as research groups and the article clusters as research topics. Visualisations are proposed to ease the interpretation of such clusters in terms of discovering leaders, the activity level, and other semantic aspects. We discuss the process of obtaining and preprocessing the information from scientific publications, the formulation and implementation of the clustering algorithm, and the creation of the visualisations. Experiments on a test data set are presented, using an initial prototype implementation of the proposed modules.

Augmented Reality for Green Consumption: Using Computer Vision to Inform the Consumers at Time of Purchase

Augmented-reality (AR) interfaces are receiving growing attention due to their versatility and usefulness in numerous application areas. In this paper, we tackle the problem of environmental awareness in consumers at the time of purchase: we design, implement, and evaluate a novel interface for overlaying product ecological information in the consumers field of vision. The identification of the product is done by computer-vision techniques that detect logotypes of brands as well as ecological labels (such as recycling symbols) when the user holds a product package. The recognition is performed with feature-detection algorithms. We evaluate the interface in terms of computational load for image processing and usability, reporting favorable results in terms of computation time, effect on the ecological consciousness of the users, and the usability.