Cifar-10-

Cifar-10 object recognition challenge with simple CNN (89% accuracy after 100 epochs). Techniques that helped increase accuracy (77%->89%):-

1. Adding dropout layers
2. Increasing the dropout value as we approach the final layer
3. Adding batch-normalisation
4. Training for a large number of epochs( Was initially training for 50 epochs)
5. Using data augmentation

Notebook 2 - Uses Transfer Learning (Resnet-34) achieves about ~91% accuracy