Bilevel optimization has been widely applied to many machine learning problems such as hyperparameter optimization, policy optimization and meta learning. Although many bilevel optimization methods recently have been proposed to solve the bilevel optimization problems, they still suffer from high computational complexities and do not consider the more general bilevel problems with nonsmooth regularization. In the paper, thus, we propose a class of enhanced bilevel optimization methods by using Bregman distance to solve bilevel optimization problems, where the outer subproblem is nonconvex and possibly nonsmooth, and the inner subproblem is strongly convex. Specifically, we propose a bilevel optimization method based on Bregman distance (BiO-BreD) for solving deterministic bilevel problems, which reaches a lower computational complexity than the best known results. Meanwhile, we also propose a stochastic bilevel optimization method (SBiO-BreD) to solve stochastic bilevel problems based on stochastic approximated gradients and Bregman distance. Moreover, we further propose an accelerated version of SBiO-BreD method (ASBiO-BreD) by using the variance-reduced technique, which achieves a lower computational complexity than the best known computational complexity with respect to condition number $\kappa$ and target accuracy $\epsilon$ for finding an $\epsilon$-stationary point. We employ data hyper-cleaning task to demonstrate that our algorithms outperform the existing bilevel algorithms.
翻译:双级优化被广泛应用于许多机器学习问题,如超参数优化、政策优化和元学习。尽管最近提出了许多双级优化方法以解决双级优化问题,但它们仍然受到高计算复杂性的困扰,没有考虑非光滑正规化的更一般双级问题。因此,在论文中,我们提议了一类强化双级优化方法,通过使用Bregman距离解决双级优化问题,即外部子问题不是混凝土,也可能是非摩擦,而内部子问题是非常强烈的。具体地说,我们提议了一种基于Bregman距离(BiO-BreD)的双级优化方法,用于解决确定性双级优化问题,其计算复杂性比已知最佳结果要低。同时,我们还提议了一种双级双级优化方法(SBiO-BreD)解决双级双级问题,以基于直观的梯度梯度梯度梯度和内部问题解决双级问题,我们进一步提议一种基于BiO-BreD值距离(BIO-BreD)的双级优化优化精度优化精度优化方法,以我们所知道的精度计算精度的精度的精度数据来显示的精度标准的精度数据,用一个已知的精度的精度的精度计算方法来显示的精度的精度的精度的精度。