On-line learning algorithm for organizing data based on generalized entropy

从各个领域收集的数据量呈指数级增长，从数据中提取有用信息的任务也变得越来越困难。为了克服计算资源的限制所带来的困难，我们研究了基于伪熵函数的从海量数据中提取有用信息的各种方法。研究结果包括以下算法：基于信息增强的自顶向下决策树学习算法；一种获取决策树近最佳剪枝的算法均匀分布下单调对数项DNF公式的学习算法。

The amount of data collected from various fields is growing exponentially and the task of exracting useful information from data is becoming more and more difficult accordingly. To overcome the difficulty that comes from the limitation on computational resources, we investigate various methodologies of exracting useful information from huge data by organizing consice data based on pseudo-entropy function. The results we obtained includes the following algorithms : top-down decision tree learning algorithm based on information based boosting ; an algorithm to obtain the nearly best prunnig of a decision trre ; an algorithm to learn monotone log-term DNF formulas under uniform distribution.