| Details: |
Graphons are limits of sequences of dense graphs. They can be used to construct latent variable models for networks. The well-known stochastic blockmodel is one example of such a model. In this talk, we will discuss various existing methods of estimating (smooth) graphons based on observed network data (such as universal singular value thresholding, neighbourhood smoothing, blockmodel approximation, matrix completion etc.) and consider the problem of extending these methods to the case where a network is not fully observed—only some overlapping subgraphs are observed. This is based on ongoing joint work with Sayak Chakrabarti, who is an undergraduate student at IIT Kanpur. |