arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1806.01756cs.AI
收藏

面向概念的深度学习

Concept-Oriented Deep Learning

IBM(国际商业机器公司)

Daniel T Chang

摘要:

概念是人类进行深度学习、理解以及知识整合和转移的基础。我们提出了面向概念的深度学习(CODL),它扩展了(机器)深度学习的概念表示和概念理解能力。CODL解决了深度学习的一些主要限制:可解释性,可转移性,上下文适应性以及对大量标记训练数据的需求。我们讨论了CODL的主要方面,包括概念图,概念表示,概念范例,概念表示学习系统支持增量和持续学习。

英文摘要:

Concepts are the foundation of human deep learning, understanding, and knowledge integration and transfer. We propose concept-oriented deep learning (CODL) which extends (machine) deep learning with concept representations and conceptual understanding capability. CODL addresses some of the major limitations of deep learning: interpretability, transferability, contextual adaptation, and requirement for lots of labeled training data. We discuss the major aspects of CODL including concept graph, concept representations, concept exemplars, and concept representation learning systems supporting incremental and continual learning.

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。

↑