tshjustin / convnn-mnist

Convolutional Neural Network trained to classify handwritten digits using MNIST Dataset - MLDA 23

Testing with CNN

Testing convolutional neural network model for the identification of handwritten digits.

Model used

A CNN model of 2 blocks of Conv-Maxpool pair are used, with the layers having 12 kernels each. An optimizer of SGD & loss function of Cross Entropy loss is used.

Implementation

The model is implemented twice, once in Pytorch and the other in Tensorflow. Both accuracy are around the same at 98%.

More kernels are added to the layer but with no apparent increase in accuracy, but adding another convolutional layer improved the accuracy to 99%.

Demo on Render

https://convnn-mnist.onrender.com/