Learning to Develop Insight: A Sub-symbolic Approach to Learning Transfer

创造性的问题解决缺乏超出预期类比或学习迁移的计算模型。许多确实存在的计算模型利用了它们所在领域的语义。这个项目考虑了一个创造性问题解决的计算模型的子符号方法。在次符号系统中创造性问题解决的发展需要在知识表征、元学习和知识转移机制方面的创新研究，并将导致对创造力的几个方面的良好的(次符号)计算解释：类比、重新表征和洞察力。为了探索和展示一种次符号化的创造性方法，该项目实施了一个学习数据流形维度固有映射的系统，并将实施一个以里程碑为基础的元学习系统，用于知识传递机制和知识表示。该项目将针对一系列问题对该系统进行分析和实证验证。更广泛的影响：拟议的研究将使系统更有能力创造性地、类似人类地解决问题，从而使社会受益；并将通过让本科生和研究生参与研究活动以及为研究生提供教学和指导机会，将教育与研究结合起来。

Creative problem solving lacks computational models that go beyond expected analogies or learning transfer. Many of the computational models that do exist exploit the semantics of the domain in which they are demonstrated. This project considers a sub-symbolic approach to a computational model of creative problem solving. The development of creative problem solving in sub-symbolic systems requires innovative research in knowledge representation, meta-learning and knowledge transfer mechanisms and will result in a well-grounded (sub-symbolic) computational explanation for several aspects of creativity: analogy, re-representation and insight. To explore and demonstrate a sub-symbolic approach to creativity, the project implements a system for learning the mappings inherent in the dimensions of a data manifold, and will implement a landmark-based meta-learning system for knowledge transfer mechanisms and knowledge representation. The project will analyze and empirically validate the system on a suite of problems. Broader Impacts: The proposed research will benefit society by making systems more capable of creative, human-like problem solving; and will integrate education with research by involving undergraduate and graduate students in research activities and by providing teaching and mentoring opportunities for the graduate students.