Find text or
[4+3+3 Marks]
7. Find a sample neural network training code using MNIST data set in a Google Colab .ipynb
Notebook shared in the following link. Create a copy of the notebook (click File option in
the Google colab menubar, then click Save a copy in Drive option) so that you can run
the code from your local Google colab account. Make sure to run all the code cells in entire
notebook (you can click Runtime option in colab menubar and then click Run all option).
In the code, you will see that a simple convolutional neural network is trained on the minist
data set. A test data set is downloaded next and then visualized. Note that the test data
has some noise and hence when the trained model is used to predict labels for the test data,
it performs very poorly achieving only around 17% of test set accuracy. The confusion
matrix reveals more confusion among the different classes.
You will aim to improve the training process so that the model achieves better test set
accuracy. For this purpose, you can make suitable modifications to the neural network
model (do not replace the model with an entirely new one, rather make only necessary
changes). Also, you can tweak the training loss and optimizer related details. You can
tweak the number of training epochs and other hyperparameters like learning rate, etc,
In addition, you are also provided an extra data set which can also be used for training.
However, note that you cannot use the test data set itself for training.
In your solution, you should share the link to your colab notebook. You should print your
notebook as a pdf file (use File option in colab menubar, click Print and then save as pdf)
and also share the pdf file as part of your solution.
In your solution, you should also describe what modifications you employed to achieve better
performance on the test data set.
[10 Marks]
Re
Ac
PD
