Group Sparse

Group sparse methods aim to improve model efficiency and interpretability by encouraging sparsity at the group level, rather than individual features. Current research focuses on applying this principle to various machine learning models, including neural networks (particularly those with additive structures or using Low-Rank Adaptation) and optimal transport methods, often employing algorithms like iteratively reweighted least squares or augmented Lagrangian approaches. This research is significant because it enhances model performance by reducing overfitting and improving generalization, while simultaneously providing insights into feature importance and relationships, leading to more efficient and interpretable models across diverse applications.