CNN CIFAR-10

This project was done as a homework assignment for a course in Machine Learning at National University of Singapore (NUS).

Alexnet

The CNN-model implemented is compared to a pre-trained model, Alexnet. Alexnet is an award-winning CNN-model trained on >15 million images split between 22.000 categories.

Accuracy

Alexnet is a complex model with 8 layers, making it computationally inefficient to train on a laptop. My model on the other hand, has only 2 layers which reduces its accuracy but makes it faster to train.

TODO: Result after n epochs

Source

Images used in this project are from the CIFAR-10 dataset, containing 60.000 images of 10 different classes with 6.000 images of each class, with 10% randomly selected pictures being test images.

Classes in CIFAR-10 dataset

• CIFAR-10: link
• Alexnet: link