Variational Garrote for Sparse Inverse Problems

A team of researchers including Kanghun Lee, Hyungjoon Soh, and Junghyo Jo has introduced the Variational Garrote (VG), a novel probabilistic method for solving sparse inverse problems. Unlike conventional L1 regularization, VG employs variational binary gating variables to approximate L0 sparsity—a closer match to the spike-and-slab structure of truly sparse signals. The researchers constructed a unified experimental framework across multiple reconstruction tasks including signal resampling, denoising, and sparse-view computed tomography, sweeping regularization strength across wide ranges to enable consistent comparison.

Experiments revealed that VG frequently achieves 40% lower minimum generalization error and improved stability in strongly underdetermined regimes where accurate support recovery is critical. The method demonstrated particular advantages in sparse-view CT reconstruction—a medical imaging technique that uses fewer X-ray projections—where VG's probabilistic approach better handles information bottlenecks. These results suggest that sparsity priors closer to spike-and-slab structure provide significant advantages when the underlying coefficient distribution is strongly sparse, highlighting the importance of prior-data alignment in inverse problems.

The study provides empirical insights into the behavior of variational L0-type methods across different information bottlenecks, showing characteristic bias-variance tradeoff patterns across tasks. By comparing VG against traditional L1 regularization through train-generalization error curves, the researchers demonstrated that VG's probabilistic gating mechanism offers superior performance when reconstruction depends on accurately identifying which signal components are truly non-zero. This work advances both theoretical understanding and practical applications of sparse regularization techniques.