【摘要】 研究了传统相关向量机(RVM)的性能,分析了传统RVM的性能完全取决于先验假设的连接权值和参数的平滑性,因而其稀疏性实际上仍受核函数或核参数选择的控制,这在某些情况下可能会导致严重的欠拟合或过拟合现象的问题,在此基础上,提出了明确地给出基函数优化过程中的目标数量,并通过最小化训练阶段前向"假定"概率分布和测试阶段反向"真实"概率分布间的交叉熵来构建RVM的方法。实验结果表明,这种方法不但可以构建最小复杂度的基于最小交叉熵的RVM结构,而且构建的RVM能很好地对数据进行拟合,提高预测的准确性,增强其稀疏性。更多还原

【Abstract】 The performance of the classical relevance vector machine(RVM) was studied,and it was analyzed that the performance of the original RVM purely depends on the smoothness of the presumed prior of the connection weights and parameters,consequently its sparsity is actually still controlled by the choice of kernel functions or kernel parameters,leading to severe underfitting or overfitting in some cases,and based on these,the RVM based on cross entropy minimization was constructed by explicitly involving the number of basis functions into the objective of the optimization procedure,and by the minimization of the cross entropy between the "hypothetical" probability distribution in the forward training pathway and the "true" probability distribution in the backward testing pathway.The experimental results show that the proposed methodology can achieve the construction of the structure with the least complexity,and the constructed RVM has the good data fitting,the good detection precision and the good sparsity.更多还原