Learning a graph topology to reveal the underlying relationship between data entities plays an important role in machine learning. Under the assumption that structured data vary smoothly over a graph, the problem can be formulated as a regularised convex optimisation over a positive semidefinite cone and solved by iterative algorithms. Classic methods require an explicit convex function to reflect the topological priors, e.g. the lasso term for sparsity, which limits the flexibility and expressi…