Group Sparsity

Group sparsity focuses on finding solutions where variables are not only individually sparse (many are zero), but also exhibit structured sparsity, meaning that entire groups of variables are simultaneously zero or non-zero. Current research emphasizes developing efficient algorithms, often based on iterative thresholding or augmented Lagrangian methods, to solve the resulting optimization problems, particularly within the context of deep learning models and dictionary learning. This approach is proving valuable in diverse applications, including adversarial attack detection, image processing, and multi-task learning, by improving model interpretability, reducing computational complexity, and enhancing performance.