用外积和法设计的权矩阵，不能保证p对模式全部正确的联想。若对记忆模式对加以限制（即要求p个记忆模式Xk是两两正交的），则用外积和法设计的BAM网具有较好的联想能力。 在难以保证要识别的样本（或记忆模式）是正交的情况下，如何求权矩阵，并保证具有较好的联想能力？这个问题在用BAM网络实现对字符的识别程序仿真中得到体现。我们做过尝试，用伪逆法求权矩阵，虽然能对未加干扰的字符全部进行识别，但对加有噪声的字符识别效果很差。至于采用改变结构和其他算法的方法来求权矩阵，将是下一步要做的工作。-foreign plot and the design of the power matrix, p is no guarantee that all the correct pattern association. If memory model, the limit (that is, p-memory model Xk is orthogonal to the February 2), then foreign plot and design of the BAM network has good ability to think. It is difficult to ensure the samples to identify (or memory mode) is orthogonal circumstances, the right to seek ways matrix, and to ensure that the association has good ability? The problem with the BAM network of characters identification procedures simulation can be manifested. We did try to use pseudo- inverse matrix for the right, although they would not increase the interference of the characters in the identification of all, However, a pair of noise increases the effects of poor character recognition. As for the