High-dimensional covariance estimation by minimizing ℓ1-penalized log-determinant divergence | Synapse