Rapid growth in networks is leading to a colossal amount of connections and data underneath.
Analysis of network measurements and structures in terms of distances, network features, etc., helps extract patterns that are not in general noticed by humans. Ideally, one should have complete set of measurements associated with a network for unbiased results. However, computational, communication, and storage limitations do not allow that.
Thus, a network is often sampled for its nodes or substructures in order to facilitate analysis. While extensive sampling can be computationally expensive for analysis or even unviable due to access restrictions, privacy issues, etc., sparse or locally targeted sampling techniques do not allow recovery of the complete information for the original network while missing facts may lead to altered network characteristics and biased research results.
Our network recovery techniques accurately predict missing data with low error values while preserving original network characteristics. We also developed ways to represent and store networks in coherent yet compressed fashion that preserve the original network characteristics for uncompromised network experiments.
This research is applicable to general graphs and data that can be represented as graphs, is scalable, and help bridge the gap between network sampling and meaningful research deductions for real-world networks.
We are currently working on:
