An efficient graph structure inference strategy based on random walk model on graph: Application to functional brain networks

A new strategy to solve the graph structure estimation problem from a noisy observed dependency matrix is proposed in this paper. It relies on an efficient inversion of the random walk model on graph employed to model the latter matrix. To this end, the sparsity assumption of the graph Laplacian associated to the target graph structure is optimally dealt with during the optimization process. The efficiency of the proposed algorithm is confirmed using both synthetic data and also using real electroencephalographic (EEG) signals in the context of inferring functional brain network.