For many important network types, physical coordinate systems and distances are either difficult to discern or inapplicable. Accordingly, characterizations based on hop-distance measurements, such as Topology Preserving Maps (TPMs) Virtual-Coordinate (VC) attractive alternatives geographic coordinates for algorithms. We present an approach recover geometric topological properties of a with small set distance measurements. The is combination shortest path (often called geodesic) recovery...

In this paper, we demonstrate how deep autoencoders can be generalized to the case of inpainting and denoising, even when no clean training data is available. particular, show neural networks trained perform all these tasks simultaneously. While, implemented by way have demonstrated potential for denoising anomaly detection, standard drawback that they require access training. However, recent work in Robust Deep Autoencoders (RDAEs) shows eliminate outliers noise a dataset without any data....

Extracting connectivity information in massive social networks is important for many applications. Algorithms developed undirected cannot be used with characterized by directed edges. We present a method to extract the network topology from small sample of distance measures without need exhaustive measurements. Tolerating missing data also necessary when certain nodes participate message passing, but are not accessible direct An anchor-based sampling approach proposed and compared random...

For many important network types (e.g., sensor networks in complex harsh environments and social networks) physical coordinate systems Cartesian), distances Euclidean), are either difficult to discern or inappropriate. Accordingly, Topology Preserving Maps (TPMs) derived from a Virtual-Coordinate (VC) system representing the distance small set of anchors is an attractive alternative coordinates for algorithms. Herein, we present approach, based on theory low-rank matrix completion, recover...

Analysis of large-scale networks is hampered by limited data as complete network measurements are expensive or impossible to collect. We present an autoencoder based technique paired with pretraining, predict missing topology information in ultra-sparsely sampled social networks. Randomly generated variations Barabási-Albert and power law cluster graphs used pretrain a Hadamard Autoencoder. Pretrained neural then infer distances where only very small fraction intra-node available. Model...

Virtual Coordinate (VC) based algorithms do not use physical coordinates for addressing, and thus possess many advantages large scale sensor networks. They rely on the validity of VCs nodes. are affected by events such as node failures which unpredictable inevitable in WSNs. This degrades performance may even reduce overall life network. A distributed algorithm is presented fault detection recovery due to failure. Consistency used detect recovered without need regenerate flooding entire...

Real-world networks have millions of users and are complex in structure. Efficient techniques required to capture the network characteristics as compact data. The overall purpose this research is mine information from partially or completely available graph data obtain optimum representation for networks. first phase involves studying draw meaningful relations properties about network. We extend work introduce a novel way sampling graphs lossless manner. second observing processing partial...

In this paper, we propose a novel method to make distance predictions in real-world social networks. As predicting missing distances is difficult problem, take two-stage approach. Structural parameters for families of synthetic networks are first estimated from small set measurements network and these then used pre-train the predictive neural Since our model searches most suitable graph which can be as an "oracle" create arbitrarily large training data sets, call approach "Oracle Search...

In this paper, we propose a novel method to make distance predictions in real-world social networks. As predicting missing distances is difficult problem, take two-stage approach. Structural parameters for families of synthetic networks are first estimated from small set measurements network and these then used pre-train the predictive neural Since our model searches most suitable graph which can be as an "oracle" create arbitrarily large training data sets, call approach "Oracle Search...