File Exchange

image thumbnail


version 1.0 (3.02 MB) by cui
Deep learning principle practice, based on MNIST hadwritten digit recognition


Updated 14 May 2020

GitHub view license on GitHub

This blog takes the classic MNIST handwritten digit recognition as an example, and gradually implements a general deep learning network model architecture step by step, without calling any third-party libraries and frameworks, using matlab for rapid construction, training, and testing. The theoretical knowledge and variable names used in the program are strictly in accordance with the symbols and formulas of the two blogs, the back update (BP) of the DNN neural network and the back propagation algorithm of the convolution neural network (CNN). MNIST handwritten digits contain 60,000 training images, 10,000 test images, the image size is 28 × 28, grayscale images, the official website gives four binary storage files, training and test data sets and label files, respectively. Assuming that the reader already understands the theoretical knowledge of the linked blog (if you are unclear, you can refer to more of the literature and the link given in the program code), we next carry out the following specific implementation.
本篇博客以经典的MNIST手写数字识别为例,逐步一步步实现通用的深度学习网络模型架构,不调用任何第三方库和框架,使用matlab进行快速搭建、训练和测试。程序中所涉及的理论知识及使用的变量名严格按照DNN神经网络的反向更新(BP)、卷积神经网络(CNN)反向传播算法 这两篇博客的符号和公式进行。MNIST手写数字包含60000张训练图片,10000张测试图片,图片大小为28×28,灰度图像,官网给出的是四个二进制存储的文件,分别为训练和测试的数据集和标签文件。假设读者已经明白所给链接博客的理论知识(不清楚可以参考更多文后的文献和程序代码中给的链接),我们接下来进行下面的具体实现。

Cite As

cui (2020). DeeplearningPractice (, GitHub. Retrieved .

Comments and Ratings (1)


Video theory explanation source:
Convolutional Neural Network in Matlab

MATLAB Release Compatibility
Created with R2006a
Compatible with any release
Platform Compatibility
Windows macOS Linux

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!