# mnist-task

**Repository Path**: l-cr/mnist-task

## Basic Information

- **Project Name**: mnist-task
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2023-10-23
- **Last Updated**: 2023-11-21

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README



## 方法

权重参数初始化（高斯分布），2个全连接层，Sigmoid，SGD，



## 实验

学习率设置为0.1，训练轮数 iters_num 设置为10000次，batch_size设置为100，

每隔 train_size / batch_size = 10000 / 100 = 600 次打印训练精度和测试精度，总共打印 iters_num / (train_size / batch_size) = 16 次



## 结果



## 参考文献

Mnist数据集原始官网

[1] Y.LeCun,“TheMNISTdatabaseofhandwrittendigits,” http://yann.lecun.com/exdb/mnist

计算图（误差反向传播法）

[2] Andrej Karpathy’s blog “Hacker’s guide to Neural Networks”.

权重参数的初始值（高斯分布）

[3] Xavier Glorot and Yoshua Bengio（2010）: Understanding the difficulty 

of training deep feedforward neural networks. In Proceedings of the 

International Conference on Artificial Intelligence and Statistics 

(AISTATS2010). Society for Artificial Intelligence and Statistics.