Data files needed to run reactive Pluto notebook to train image classifier

Two CSV files with 10,000 random training and test samples each of MNIST data (28 x 28 images of handwritten digits). These files are required to reproduce the results of Experiment 3 in the article "Constructing structured tensor priors for Bayesian inverse problems" via a reactive Pluto notebook. In this experiment a classifier is learned via Bayes' Theorem for images of 10 handwritten digits (0 to 9). The notebook illustrates the use of different structured tensor priors and compares the performance of the resulting classifier with the standard Tikhonov prior on new test data.本数据集包含两份CSV文件，每份均包含10000条随机采样的训练样本与测试样本，对应MNIST手写数字数据集（由28×28像素的手写数字灰度图像构成）。借助交互式Pluto笔记本，可复现论文《为贝叶斯逆问题构建结构化张量先验》（Constructing structured tensor priors for Bayesian inverse problems）中实验3的完整结果。该实验基于贝叶斯定理（Bayes' Theorem）针对0至9共10类手写数字图像训练分类器，本交互式Pluto笔记本将演示不同结构化张量先验（structured tensor priors）的具体应用，并在全新测试数据集上对比基于各类先验训练得到的分类器与标准吉洪诺夫先验（Tikhonov prior）分类器的性能表现。