Sub-optimality of the Naive Mean Field approximation for proportional high-dimensional Linear Regression

Despite its popularity in practice, theoretical guarantees for high-dimensional problems are only available under strong structural assumptions (e.g., sparsity).