Efficient Algorithms for Learning and Reasoning from Data

点击翻译按钮获取中文摘要

The research has as its goal the development of efficient, scalable and portable algorithms that allow learning and limited reasoning over very large repositories of data. The goal is to combine these algorithms into an environment that supports summarization of very large streams of data (data mining) and the drawing of reasonably accurate conclusions from these summaries, using probabilistic reasoning where appropriate. The following areas are being investigated: (1) Efficient algorithms for reasoning in probabilistic networks; (2) Efficient induction of provably accurate generalized decision trees; and (3) Incorporating knowledge in memory-based reasoning methods. Of particular interest are reasoning algorithms that perform queries and updates to data extremely efficiently, perhaps in time which is sublinear to the size of the original database, and learning algorithms that provide performance guarantees on the accuracy of the answers by giving a confidence rating with the answer to a query. The availability of high-speed, reliable, intelligent algorithms for learning and reasoning from data will have important applications in digital libraries, medical care, process control, and resource allocation, among other areas.