Handwritten Digit Recognition

 

For this project I used the MNIST Handwritten Digits dataset which is commonly used in computer vision and deep learning. I built a Multilayer Perceptron and used the Relu activation function for my hidden layers and Softmax Activation function for the output layer.
After training the model on 50 epochs with a learning rate of 1e-3 I achieved a training accuracy of 99.2% and an accuracy of 97.84% on the test dataset. Furthermore, I used Name Scoping and Image Visualization in Tensorboard to group nodes and ensure better visualization.

The jupyternotebook, histograms and graphs are attached below and here is the link to my Github.


Jupyter Notebook


Accuracy and loss accuracy-and-loss


Main Graph graph-run-512-DO-64-LR0


Histograms for all the layers layer-1-histogram layer-2-histogram out-histogram