Efficient and Accelerated Online Learning for Sparse Group Lasso

The batch-mode group lasso algorithms suffer from the inefficiency and poor scalability, and online learning algorithms for group lasso, is a promising tool for attacking the large-scale problem. However, the low time complexity of current online algorithm often be accompanied by low convergence rate, and the faster convergence rate is a key problem to guarantee the online learning algorithms. We develop a novel accelerated online learning algorithm to solve sparse group lasso model. The sparse group lasso model can achieve more sparsity in both the group level and the individual feature level. By adopting dual averaging method, its worst-case time complexity and memory cost at each iteration are both in the order of O (d), where d is the number of dimensions. Moreover, our online algorithm has a accelerated capability, and its theoretical convergence rate is O (1/T2) up to T-th step. The experimental results on synthetic and real-world datasets demonstrate the merits of the proposed online algorithm for sparse group lasso.