Strength from Weakness: Fast Learning Using Weak Supervision
Strength from Weakness: Fast Learning Using Weak Supervision
复制标题
DOI：
--
复制
发表时间：
2020-02
期刊：
影响因子：
--
通讯作者：
Joshua Robinson;S. Jegelka;S. Sra
中科院分区：
文献类型：
--
作者：
Joshua Robinson;S. Jegelka;S. Sra
我们研究了弱监督学习的泛化性质。也就是说，学习只有少数“强”标签（我们预测的实际目标）存在，但更多的“弱”标签可用。特别是，我们表明，访问弱标签可以显着加快强任务的学习速度，达到$\mathcal{O}（\nicefrac1n）$的快速速度，其中$n$表示强标签数据点的数量。即使强标记数据本身只允许较慢的$\mathcal{O}（\nicefrac{1}{\sqrt{n}}）$速率，这种加速也会发生。实际的加速持续依赖于可用的弱标签的数量，以及两个任务之间的关系。我们的理论结果在一系列任务中得到了经验性的反映，并说明了弱标签如何加速强任务的学习。
We study generalization properties of weakly supervised learning. That is, learning where only a few "strong" labels (the actual target of our prediction) are present but many more "weak" labels are available. In particular, we show that having access to weak labels can significantly accelerate the learning rate for the strong task to the fast rate of $\mathcal{O}(\nicefrac1n)$, where $n$ denotes the number of strongly labeled data points. This acceleration can happen even if by itself the strongly labeled data admits only the slower $\mathcal{O}(\nicefrac{1}{\sqrt{n}})$ rate. The actual acceleration depends continuously on the number of weak labels available, and on the relation between the two tasks. Our theoretical results are reflected empirically across a range of tasks and illustrate how weak labels speed up learning on the strong task.