STTR Phase I: Disciple Technologies for Development, Utilization, and Maintenance of Regulatory Knowledge Bases

这个小企业技术转移研究第一阶段项目将进行可行性研究，使用学徒学习和互动推理方法，以及乔治梅森大学学习代理中心开发的门徒技术，为动态金融服务组织建立、利用和维护监管知识库。弟子学习软件代理可以由专家直接教授，成为知识助手。专家直接与门徒代理互动，教它解决问题，就像专家教人类学徒一样，通过给代理示例和解释，以及监督和纠正其行为。智能体从专家那里学习，通过归纳例子和解释来建立知识库。弟子已被应用于军事挑战问题和情报分析。STTR第一阶段工作的目标是使用Disciple技术为金融服务公司构建原型监管知识库。该原型将有助于对这些类型的应用领域的门徒技术进行可行性分析。每一家现代企业不仅要应对本地和全球的竞争对手，还要应对本地和全球的法规。这些法规给企业运营带来了巨大的负担，导致员工和信息系统必须遵守存储在大量不断变化的文件中的许多政策。当前的知识管理技术和软件工具允许有效地组织和检索文档，但是它们不能提供有效的定制过程来处理特定的业务用例。因此，需要像Disciple这样灵活的知识系统，在遵守所有规章制度的同时，为解决具体案件提供帮助。这些系统还应该能够获得其用户的推理技能，以调整其处理新情况的能力。基于第一阶段构建的原型，将评估Disciple支持开发、利用和维护监管知识库的能力。原型还将在以下领域帮助确定第二阶段的研究和开发目标：定量推理和学习能力、知识库版本控制、共享和审计功能，以及与数据库引擎的接口。

This Small Business Technology Transfer Research Phase I project will perform feasibility studies of using apprenticeship learning and interactive reasoning methods, and the Disciple Technologies developed by the George Mason University Learning Agents Center, to build, utilize, and maintain regulatory knowledge bases for dynamic financial services organizations. The Disciple learning software agent can be taught directly by an expert to become a knowledgebased assistant. The expert interacts directly with a Disciple agent to teach it to solve problems in a way that is similar to how the expert would teach a human apprentice, by giving the agent examples and explanations, as well as by supervising and correcting its behavior. The agent learns from the expert by generalizing the examples and the explanations to build its knowledge base. Disciple has been applied to military challenge problems and to the intelligence analysis. The objective of this STTR Phase I effort is to use the Disciple technologies to build a prototype regulatory knowledge base for a financial services firm. The prototype will help to conduct a feasibility analysis of the Disciple technologies for these type of application domains. Every modern enterprise must deal not only with local and global competitors, but also with local and global regulations. These regulations put tremendous burden on business operations causing workers and information systems to comply with many policies stored in a large number of ever changing documents. The current knowledge management technologies and software tools allow efficient organization and retrieval of documents, but they cannot offer efficient customized procedures to deal with specific business cases. Therefore, there is a need for flexible knowledge-based systems, like Disciple, that will offer help in solving specific cases while complying with all the rules and regulations. These systems should also be capable of acquiring reasoning skills of their users to adapt their capabilities to deal with new cases. Based on theprototype built during Phase I, Disciple abilities to support development, utilization, and maintenance of regulatory knowledge bases will be evaluated. The prototype will also help to identify research and development goals for Phase II in the following areas: quantitative reasoning and learning capabilities, knowledge base versioning, sharing, and auditing functions, and interfacing with database engines.