We use bilevel optimization to tune the model’s hyperparameters. 我们使用双层优化来调整模型的超参数。
Bilevel optimization formulates training as an upper-level objective while the lower-level problem enforces that the network parameters minimize the training loss under given regularization. 双层优化把训练表述为上层目标,同时用下层问题约束网络参数:在给定正则化条件下使训练损失最小。
Dempe, S.Foundations of Bilevel Programming(系统介绍双层规划的理论基础与典型问题)
Dempe, S.; Zemkoho, A.Bilevel Optimization: Advances and Next Challenges(双层优化的研究进展与挑战综述)
Colson, B.; Marcotte, P.; Savard, G. “An overview of bilevel optimization”(关于双层优化的经典综述性论文)
Franceschi, L. et al. “Bilevel Programming for Hyperparameter Optimization and Meta-Learning” / “A Bilevel Optimization Approach for Hyperparameter Optimization”(机器学习中用双层优化做超参数与元学习的代表性工作)