Abstract:In recent decades, large scale of data sets have been accumulated in various and wide fields, in particular in social and biological systems. There are massive amounts of data available for utilization; and for really using these data it is crucial to find effective methods of analyses. Many practical systems can be described by dynamic networks, for which modern technique can measure their outputs, and accumulate extremely rich data. Nevertheless, the network structures producing these data are often deeply hidden. The structure and intensity of couplings between the network elements (say, network nodes) play key roles in determining the features of network dynamics. Therefore, the problem of network reconstruction, i.e., exploring unknown network structures by analyzing available output network data, has attracted significant interest in many interdisciplinary fields in recent decades. The significance of network reconstruction is due to not only the extreme importance of black-box manifestation in practice, but also serious difficulties, which are theoretically challenging, and appearing in various aspects of real-world systems, such as complexity of network structures, strong nonlinearity of network dynamics, diverse and unknown impacts from the interiors of nodes and the externals of networks, i.e., presence of noises, many hidden and not measurable nodes in networks. In this review, we introduce some effective methods dealing with the above difficulties, such as exploring information of available data by searching rich and diverse data features; seeking different types of correlations aiming at treating different difficulties; finding intelligent ways to remove negative and utilize active effects of noises, and so on. Our approach offers some possibilities towards understanding of complex networked systems in more complicated and difficult conditions and has important implications for the reconstruction tasks of many practical networks.

Unsupervised Learning from Noisy Networks with Applications to Hi-C Data.

Locating High-Density Clusters with Noisy Queries.

Unsupervised Hierarchical Feature Selection on Networked Data.

From data to network structure-Reconstruction of dynamic networks

From Data to Network Structure&mdash;&mdash;reconstruction of Dynamic Networks

Learning Node Representations from Noisy Graph Structures

Reconstruction of Noise-Driven Nonlinear Networks from Node Outputs by Using High-Order Correlations

GNN Cleaner: Label Cleaner for Graph Structured Data

Depicting Network Structures from Variable Data Produced by Unknown Colored-Noise Driven Dynamics

A Learning Framework to Improve Unsupervised Gene Network Inference

Unsupervised Learning for Community Detection in Attributed Networks Based on Graph Convolutional Network

A Unified Framework for Exploratory Learning-Aided Community Detection Under Topological Uncertainty

Network Enhancement: a general method to denoise weighted biological networks

Active Learning for Graphs with Noisy Structures

Network enhancement as a general method to denoise weighted biological networks

Unsupervised Instance and Subnetwork Selection for Network Data

Unsupervised Network Embedding for Graph Visualization, Clustering and Classification

An Incremental Network for On-Line Unsupervised Classification and Topology Learning.

Uncovering hidden nodes and hidden links in complex dynamic networks

Network Structure-Enhanced Extremal Optimization Based Semi-supervised Algorithm for Community Detection

A sampling-guided unsupervised learning method to capture percolation in complex networks