arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2006.08791cs.LGstat.ML
收藏

间接监督信号下的可学习性

Learnability with Indirect Supervision Signals

University of Pennsylvania(宾夕法尼亚大学);Allen Institute for AI(艾伦人工智能研究所);University of Illinois at Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校);Amazon(亚马逊)

Kaifu Wang, Qiang Ning, Dan Roth

摘要:

从间接监督信号中学习在现实世界的人工智能应用中非常重要,因为通常情况下,黄金标签缺失或成本过高。在本文中,我们开发了一个统一的多类分类的理论框架时,监督是由一个变量,包含非零的互信息的黄金标签。这个问题的性质是由(i)从黄金标签到间接监督变量的转换概率和(ii)学习者关于转换的先验知识决定的。我们的框架放宽了文献中的假设,并支持学习未知的,不可逆的和依赖实例的过渡。我们的理论引入了一个新的概念称为\emph{separation},它的特点是可学习性和推广的界限。我们还展示了我们的框架的应用程序,通过具体的新颖的结果,在各种学习场景,如学习超集注释和联合监督信号。

英文摘要:

Learning from indirect supervision signals is important in real-world AI applications when, often, gold labels are missing or too costly. In this paper, we develop a unified theoretical framework for multi-class classification when the supervision is provided by a variable that contains nonzero mutual information with the gold label. The nature of this problem is determined by (i) the transition probability from the gold labels to the indirect supervision variables and (ii) the learner's prior knowledge about the transition. Our framework relaxes assumptions made in the literature, and supports learning with unknown, non-invertible and instance-dependent transitions. Our theory introduces a novel concept called \emph{separation}, which characterizes the learnability and generalization bounds. We also demonstrate the application of our framework via concrete novel results in a variety of learning scenarios such as learning with superset annotations and joint supervision signals.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑