How can we employ theoretical insights and practical tools from knowledge representation and reasoning to enhance machine learning, and when is it worthwhile to do so? This paper is based on an invited talk delivered at ECSQARU2019 around this question. It emphasizes the knowledge representation and reasoning side of knowledge-enhanced machine learning, looking at a few case studies: the finite model theory of probabilistic languages, the generation of explanations for embeddings, and an “explainable” version of the Winograd Challenge.

中文翻译：

---

关于知识增强型机器学习的一些思考

我们如何利用知识表示和推理的理论见解和实践工具来增强机器学习，何时值得这样做？本文基于在 ECSQARU2019 上围绕这个问题发表的受邀演讲。它强调了知识增强型机器学习的知识表示和推理方面，着眼于一些案例研究：概率语言的有限模型理论、嵌入解释的生成以及 Winograd 挑战的“可解释”版本。