CNN on MNIST Data

---

Link to colab : https://colab.research.google.com/drive/1qPKo-Ys5iYg5Pon646tUGSg289I4ZbYn

Target:

Add Image Augmentation to get a test accuracy target of 99.4% or above. I used RandomAffine which consiste of slight rotation, translation and scaling based on image analysis from the first file.

Results:

Parameters: 10,008

Best Train Accuracy: 98.57%

Best Test Accuracy: 99.40%

Analysis:

Despite adding dropout, we still see some overfitting because there are already very few parameters. So, I used a dropout of 0.

I used random affine as data transformation which consist of slight rotation, translation and scaling. Image augmentation increased the test accuracy to 99.4%.