---
title: Model_1
canonical_url: "https://www.modelscope.cn/models/RakiSta/Model_1"
md_url: "https://www.modelscope.cn/models/RakiSta/Model_1.md"
repository: RakiSta/Model_1
chinese_name: "CS30064 PJ1 Models"
last_updated: 2026-05-19
pipeline_tag: ocr-recognition
tasks:
  - ocr-recognition
library_name:
  - pytorch
frameworks:
  - Pytorch
downloads: 1
stars: 0
---

# Model_1

> Model_1 - RakiSta 在 ModelScope 开源的模型。Model 14 for Handwritten Digit Classification on MNIST Dataset

- **Repository**: RakiSta/Model_1
- **Tasks**: ocr-recognition
- **Downloads**: 1
- **Stars**: 0
- **Last updated**: 2026-05-19

Source: https://www.modelscope.cn/models/RakiSta/Model_1

---

# Model 1~4 for Handwritten Digit Classification on MNIST Dataset

## Introduction

best_models_1 (MLP), best_models_2 (CNN), best_models_3 (CNN) and best_models_4 (CNN) are the models trained and used in the project for the course "CS30064 Neural Networks and Deep Learning" of Fudan University. The goal of this project is to investigate handwritten digit classification on the MNIST dataset using MLP and CNN models with different settings.

## Model Architecture

The MLP model is a simple feedforward neural network with one hidden layer. The input layer has 784 nodes (corresponding to the 28x28 pixel input images), and the output layer has 10 nodes (corresponding to the 10 possible digit classes). The hidden layer has 600 nodes and the ReLU activation function is used.

The CNN models is a convolutional neural network with two convolutional layers and two fully connected layers. The input layer has 1 channel (corresponding to the grayscale input images), and the output layer has 10 nodes (corresponding to the 10 possible digit classes). The only convolutional layer has 16 filters of size 3x3, and the ReLU activation function is used. The fully connected layer has 10816 input nodes and 10 output nodes.

## Model Training

- Loss funtion: cross-entropy loss function
- Optimizer:
  - Model 1 and Model 2: SGD optimizer with a learning rate of 0.05
  - Model 3: SGD optimizer with a learning rate of 0.2
  - Model 4: Momentum optimizer with a learning rate of 0.05 and a momentum of 0.9
- Epochs: 20
- Batch size: 1024
- Evaluation metric: accuracy

## Model Evaluation

According to the results of the model evaluation, the accuracy of 4 models on the test set (10000 images of MNIST dataset) are:

- Model 1: 92.91%
- Model 2: 95.43%
- Model 3: 97.92%
- Model 4: 98.23%
