ITR: Knowledge-Enhanced Discovery System (KEDS): Incorporating Background Knowledge for Scientific Discovery

KEDS项目正在开发一个用于科学发现的工具包，它指的是从有关这些现象的观测数据中为一系列感兴趣的现象确定新的预测模型。KEDS超越了现有的方法，将背景知识（如变量之间的依赖关系、部分模型和关于学习模型将用于的任务的信息）纳入学习过程，并提供分析工具来帮助用户理解、评估和比较学习模型。KEDS为科学发现提供了一种比现有工具更灵活、更互动的方法。KEDS中的互动技术也被应用于科学教育，支持一个迭代的过程，在这个过程中，学生根据来自系统的个性化、有针对性的反馈来完善他们的心理模型。该项目的影响将是在如何在科学调查和科学教育中使用信息技术方面取得进展。特别是，KEDS将支持一种更具互动性的科学发现方式，这将允许人类领域的专家更有效地将他们以前的知识整合到发现过程中。预期的结果包括改进的科学发现和科学教育的交互式方法，交互式科学发现的数据集和基准，以及一个文档化的软件包。KEDS系统目前正应用于天文科学领域，但其基础技术也广泛适用于许多其他科学领域，包括地球科学、生物和医学、化学和材料科学。

The KEDS project is developing a toolkit for scientific discovery, which refers to the identification of new predictive models for a set of phenomena of interest from observational data about those phenomena. KEDS goes beyond existing methods by incorporating background knowledge (such as dependencies between variables, partial models, and information about the task for which the learned models will be used) into the learning process, and by providing analysis tools to aid the user in understanding, evaluating, and comparing the learned models.KEDS provides a more flexible, interactive approach to scientific discovery than current tools. The interactive techniques in KEDS are also being applied to science education, supporting an iterative process in which students refine their mental models, based on personalized, targeted feedback from the system.The impact of the project will be an advance in how information technology is used in scientific investigations and science education. In particular, KEDS will support a more interactive style of scientific discovery, which will allow human domain experts to integrate their previous knowledge into the discovery process more effectively. Expected results include improved interactive methods for scientific discovery and science education, data sets and benchmarks for interactive scientific discovery, and a documented software package. The KEDS system is currently being applied to astronomy science domains, but the underlying techniques have broad applicability to many other science domains, including earth sciences, biology and medicine, chemistry, and materials science.