Natali Ruchansky

Learn More
The introduction of Software Defined Networks (SDNs) is completely changing the way in which networks are built and managed. SDNs decouple data from control plane access, which makes introduction of new network functionalities significantly simpler. The philosophy of OpenFlow is a move towards centralization, where a single controller program manages the(More)
Consider this simple question: how can a network operator identify the set of routes that pass through its network? Answering this question is surprisingly hard: BGP only informs an operator about a limited set of routes. By observing traffic, an operator can only conclude that a particular route passes through its network -- but not that a route does not(More)
Characterizing the set of routes used between domains is an important and difficult problem. The size and complexity of the millions of BGP paths in use at any time can hide important phenomena and hinder attempts to understand the path selection behavior of ASes. In this paper we introduce a new approach to analysis of the interdomain routing system(More)
The Wiener index of a graph is the sum of all pairwise shortest-path distances between its vertices. In this paper we study the novel problem of finding a <i>minimum Wiener connector</i>: given a connected graph <i>G=(V,E)</i> and a set <i>Q</i> &#8838; <i>V</i> of query vertices, find a subgraph of <i>G</i> that connects all query vertices and has minimum(More)
In this paper, we describe a physical activity classification system using a body sensor network (BSN) consisting of cost-sensitive tri-axial accelerometers. We focus on workspace activities (different motions and sitting postures). We use a Naive Bayes classifier and show that we can train the system simply and systematically. For each task, we find a set(More)
Matrix completion is a problem that arises in many dataanalysis settings where the input consists of a partiallyobserved matrix (e.g., recommender systems, traffic matrix analysis etc.). Classical approaches to matrix completion assume that the input partially-observed matrix is low rank. The success of these methods depends on the number of observed(More)