模式识别系统中的一次性风险判据问题

摘要

The paper is concerned with making single shot decisions with arisk in the case of alternatives with negative outcomes(losses).Requirements are formulated to the criterion of choosing the least dangerous alternative. A way is suggested for estimating the danger of implementing alternatives with two possible outcomes of which one does not entail dangerous consequences.%该文讨论模式识别系统，当备选方案的损失代价相互矛盾时，进行一次性风险决策的判据问题，给出选择风险最小方案时的判据条件及风险测评方法。提出决策面谨慎度位于Neyman-pearson判决准则和极小化极大判据之间的最小风险估计函数∪（l,p,B）及其具体表达形式。通过例子说明这一函数在风险决策中的应用及其在利用计算机仿真确定分类样本最佳损失矩阵中的重要作用。