中文摘要：

为解决传统模式识别方法学习结果过于复杂且难以解读的问题，提出了一种基于遗传算法的演化学习超网络模型．与传统的基于梯度下降和超边替代的超网络学习算法不同，演化学习超网络模型在其学习过程中引入了遗传算法．将超网络的超边集合划分成多个子种群；对子种群中的个体进行选择、交叉和变异等遗传操作，并对每一代种群进行子种群间优秀个体的迁移．每个子种群并行执行演化操作，完成演化后得到一个具有决策能力的超网络分类器．利用演化超网络对急性白血病、肺癌和前列腺数据集进行分类试验．结果表明，演化学习超网络对3个数据集的分类准确率分别为96．21％，99．26％，96．09％．所提出的方法与其他传统的模式识别方法相比，具有更高的分类准确率，而且其学习结果具有很好的可读性，有利于挖掘与癌症诊断密切相关的基因对高阶关联关系．

英文摘要：

In order to solve the shortcoming of relatively complicate and barely intelligible classification result of most conventional machine learningbased pattern recognition method, an evolutionary learning hypernetwork evolved by a genetic algorithm was proposed. Differed from the traditional hypernetwork based on a gradient descent or a hyperedge replacement scheme as system learning machine, a genetic al gorithm was employed in the learning process of the proposed evolutionary learning hypernetwork. In the system learning process, the hyperedges of hypernetwork were divided into several subgroups for the in dividuals to be evolved independently with selection, crossover and mutation operations. The outstanding individuals were migrated to the neighbor subgroup in every generation. Every subgroup was evolved with genetic algorithm parallel, and eventually contributed to a hypernetworkbased classifier with the ability of decisionmaking. The evolved hypernetwork was used to classify the data set of acute leukemia, lung cancer and prostate. The experimental results show that the proposed approach leads to a very com parable classification performance with data set accuracies of 96.21% on acute leukemia, 99.26% on lung cancer and 96.09% on prostate, respectively. The learning results of the proposed hypernetwork aremore readable than those of other traditional classification methods. The proposed scheme can efficiently discover significant high order interactions of gene pairs for cancer classification.