AutoLoss: Learning Discrete Schedules for Alternate Optimization
By Haowen Xu, H. Zhang, Zhiting Hu, Xiaodan Liang, Ruslan Salakhutdinov, Eric P. Xing. 2018
30 citations recorded by the source corpus.
Many machine learning problems involve iteratively and alternately optimizing different task objectives with respect to different sets of parameters. Appropriately scheduling the optimization of a task objective or a set of parameters is usually crucial to the quality of convergence. In this paper, we present AutoLoss, a meta-learning framework that automatically learns and determines the optimization schedule. AutoLoss provides a generic way to represent and learn the discrete optimization schedule from metadata, allows for a dynamic and data-driven schedule in ML problems that involve alternating updates of different parameters or from different loss objectives. We apply AutoLoss on four ML tasks: d-ary quadratic regression, classification using a multi-layer perceptron (MLP), image generation using GANs, and multi-task neural machine translation (NMT). We show that the AutoLoss controller is able to capture the distribution of better optimization schedules that result in higher quality of convergence on all four tasks. The trained AutoLoss controller is generalizable -- it can guide and improve the learning of a new task model with different specifications, or on different datasets.
Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Orhan Firat
- Xuanyi Dong
- Melvin Johnson
- Yezhou Yang
- Graham Neubig
- Sébastien Jean
- Felix Stahlberg
- Dong Yang
In Computer Vision and Pattern Recognition
Citations in the corpus
- Searching Learning Strategy with Reinforcement Learning for 3D Medical Image Segmentation (2019, methodology): “Alternatively, RL based approaches can be applied to the tasks of optimizing other parts of neural network model training, such as data augmentation policies [4] and design of loss functions [15] in order to bypass some limitations of differentiable architecture search.”
- Adaptive Scheduling for Multi-Task Learning (2019, background): “Xu et al. present AutoLoss [20], which uses reinforcement learning to train a controller that determines the optimization schedule.”
- Teacher Supervises Students How to Learn From Partially Labeled Images for Facial Landmark Detection (2019, methodology): “In this sense, we are related to some recent literature in “learning to learn” [25, 33, 13, 45].”