Classifiers based on neural networks () often lack a measure of uncertainty in the predicted class. We propose a method to estimate the probability mass function () of the different classes, as well as the covariance of the estimated . First, a local linear approach is used during the training phase to recursively compute the covariance of the parameters in the . Secondly, in the classification phase, another local linear approach is used to propagate the covariance of the learned parameters to the uncertainty in the output of the last layer of the . This allows for an efficient Monte Carlo () approach for; (i) estimating the ; (ii) calculating the covariance of the estimated ; and (iii) proper risk assessment and fusion of multiple classifiers. Two classical image classification tasks, i.e., , and 10, are used to demonstrate the efficiency of the proposed method.

中文翻译：

---

神经网络分类器中的不确定性量化——局部线性方法

基于神经网络的分类器通常缺乏对预测类别的不确定性的测量。我们提出了一种方法来估计不同类别的概率质量函数 () 以及估计的协方差。首先，在训练阶段使用局部线性方法来递归计算 中参数的协方差。其次，在分类阶段，使用另一种局部线性方法将学习参数的协方差传播到最后一层输出的不确定性。这允许采用高效的 Monte Carlo () 方法； (i) 估计； (ii) 计算估计值的协方差； (iii) 适当的风险评估和多个分类器的融合。两个经典图像分类任务，即 和 10，被用来证明该方法的效率。