一种基于CRO的高阶神经网络多示例学习方法

在多示例学习(MIL)中,包是含有多个示例的集合,训练样本只给出包的标记,而没有给出单个示例的标记。提出一种基于示例标记强度的MIL方法(ILI-MIL),其允许示例标记强度为任何实数。考虑到基于梯度训练神经网络方法的计算复杂性和ILI-MIL目标函数的复杂性,利用基于化学反应优化的高阶神经网络来实现ILI-MIL,学习方法具有较强的非线性表达能力和较高的计算效率。实验结果表明,该算法比已有算法具有更加有效的分类能力,且适应范围更广。

Multiple-instance Learning Method Based on CRO High Order Neural Networks

Multi-instance learning (MIL) is a variant of inductive machine learning developed recently,in which each learning example contains a bag of instances instead of a single feature vector.In this paper,we presented a novel MIL method based on the concept of instance label intensity(ILI) called ILI-MIL.Considering the complexity of the object function and the complexity of the gradient descent based training method in neural networks,we used a chemical reaction optimization (CRO) algorithm for training a high-order neural network (HONN) to implement the presented ILI-MIL method,which has more powerful nonlinear fitting capacity and high computation efficiency.The experiment results show that our ILI-MIL method have more effective ability of classification than the state-of-the-art methods.