ICML 2020poster78 citations

Meta-learning for Mixed Linear Regression

Weihao Kong, Raghav Somani, Zhao Song, Sham Kakade, Sewoong Oh

Abstract

In modern supervised learning, there are a large number of tasks, but many of them are associated with only a small amount of labelled data. These include data from medical image processing and robotic interaction. Even though each individual task cannot be meaningfully trained in isolation, one seeks to meta-learn across the tasks from past experiences by exploiting some similarities. We study a fundamental question of interest: When can abundant tasks with small data compensate for lack of tasks with big data? We focus on a canonical scenario where each task is drawn from a mixture of $k$ linear regressions, and identify sufficient conditions for such a graceful exchange to hold; there is little loss in sample complexity even when we only have access to small data tasks. To this end, we introduce a novel spectral approach and show that we can efficiently utilize small data tasks with the help of $\tilde\Omega(k^{3/2})$ medium data tasks each with $\tilde\Omega(k^{1/2})$ examples.

BibTeX
@InProceedings{pmlr-v119-kong20a,
  title = 	 {Meta-learning for Mixed Linear Regression},
  author =       {Kong, Weihao and Somani, Raghav and Song, Zhao and Kakade, Sham and Oh, Sewoong},
  booktitle = 	 {Proceedings of the 37th International Conference on Machine Learning},
  pages = 	 {5394--5404},
  year = 	 {2020},
  editor = 	 {III, Hal Daumé and Singh, Aarti},
  volume = 	 {119},
  series = 	 {Proceedings of Machine Learning Research},
  month = 	 {13--18 Jul},
  publisher =    {PMLR},
  pdf = 	 {http://proceedings.mlr.press/v119/kong20a/kong20a.pdf},
  url = 	 {https://proceedings.mlr.press/v119/kong20a.html},
  abstract = 	 {In modern supervised learning, there are a large number of tasks, but many of them are associated with only a small amount of labelled data. These include data from medical image processing and robotic interaction. Even though each individual task cannot be meaningfully trained in isolation, one seeks to meta-learn across the tasks from past experiences by exploiting some similarities. We study a fundamental question of interest: When can abundant tasks with small data compensate for lack of tasks with big data? We focus on a canonical scenario where each task is drawn from a mixture of $k$ linear regressions, and identify sufficient conditions for such a graceful exchange to hold; there is little loss in sample complexity even when we only have access to small data tasks. To this end, we introduce a novel spectral approach and show that we can efficiently utilize small data tasks with the help of $\tilde\Omega(k^{3/2})$ medium data tasks each with $\tilde\Omega(k^{1/2})$ examples.}
}
Meta-learning for Mixed Linear Regression · ICML 2020