WenliangGuo / Mnist

Solving the Mnist task by constructing multiple fully-connected layers without using existing DL frameworks.

Using MLP to Recognize Handwritten Numbers

Environment

Platform: Windows / Linux
Packages: Numpy + Matplotlib

Dataset

MNIST

Usage

Running the Mnist.ipynb file

Parameters

lr: learning rate
batch: batch size
epoch: the number of training rounds
cell _ num1 / cell _ num2 / cell _ num3: the number of neurals in three hidden layers

Experiment Results

Training Loss

Minimum Loss: 0.000068

Testing Accuracy

Maximum Accuracy: 0.981500