Subdeterminant Maximization via Nonconvex Relaxations and Anti-concentration