New closed-form regularization eliminates cross-validation, speeding sparse estimation 10,000x

Sparse precision matrix estimation is critical for modeling conditional dependencies in high-dimensional, low-sample-size data, but choosing the regularization parameter that balances sparsity and fit has traditionally required costly cross-validation. In a new paper on arXiv, researchers Aryan Eftekhari, Daniel Sergio Vega, Ernst-Jan Camiel Wit, and Olaf Schenk propose a closed-form, matrix-valued regularization parameter derived from the sampling distribution of the first-order optimality conditions of the ℓ₁-regularized Gaussian maximum-likelihood estimator. By prescribing the probability that each nonzero entry satisfies its optimality condition under resampling, the method completely eliminates cross-validation.

Under standard conditions, the approach attains asymptotic scaling properties that guarantee consistency and sparsistency (correct identification of zero/nonzero entries). On synthetic Gaussian and non-Gaussian datasets as well as real-world gene microarray and neuroimaging applications, the proposed method achieves estimation accuracy comparable to cross-validation, delivers superior support recovery, and reduces runtime by several orders of magnitude. This breakthrough offers a practical, scalable solution for high-dimensional graphical model selection.