A1258
Title: An algorithm for solving the constrained sparse-group lasso
Authors: Nazgul Zakiyeva - Institute of Mathematics and Mathematical Modelling (Kazakhstan) [presenting]
Abstract: The aim is to propose a mixed coordinate descent method of multipliers for solving the constrained sparse-group lasso problem in multivariate time series analysis. The constrained sparse-group lasso extends the widely used sparse-group lasso by incorporating linear constraints, which enable the integration of prior information into the model, such as balance conditions between supply and demand. The proposed method combines coordinate descent with augmented Lagrangian techniques, yielding an efficient and flexible optimization framework that simultaneously addresses sparsity, group structure, and linear restrictions. The methodology is examined through numerical experiments and real data applications, and benchmarked against alternative optimization approaches.
