论文部分内容阅读
文章提出了模式识别的最大熵方法,其基本思想是求出最大熵概率分布,再求出条件概率分布,进而作出二值分类。它的特点是能最大限度地利用已有信息做出最合理的推测。求解最大熵分布时,需要解复杂的约束优化问题,为此使用了神经网络,从而使该方法结合了神经网络的很多优点。该方法的突出优点是在小样本情况下仍能保持很好的识别率。
In this paper, the maximum entropy method for pattern recognition is proposed. The basic idea is to find the maximum entropy probability distribution, then to find the conditional probability distribution, and then to make binary classification. It is characterized by the maximum possible use of existing information to make the most reasonable guess. When solving the maximum entropy distribution, we need to solve the problem of complicated constrained optimization. To this end, neural network is used, so that the method combines many advantages of neural network. The salient advantage of this method is that it still maintains a good recognition rate in the case of small samples.