This is a deep neural network classiifier that uses covolution neural network with 2 FC layers to classify Handwritten digits from MNIST.
MNIST dataset contains gray scale images of size 28 * 28. The train data has 60000 images and test has 10000 images. Below is the sample data
Architecure diagram is below:
- matplotlib==3.7.1
- matplotlib-inline==0.1.6
- torch==2.0.1
- torchsummary==1.5.1
- torchvision==0.15.2
- tqdm==4.65.0
To run the code, execute the Session_5.ipynb file. The model is set to execute on MPS/GPU/CPU The training time on MPS is ~ 4 mins
The model reached and accuracy of 99.28% on test set with loss of 0.01. The loss function used here was Crossentropy loss.
- Total number of trainable model parameters = 593,200
- Total number of non-trainable model parameters = 0
- Model Size = 2.94MB