tinanemati / fashion_mnist

Created a neural network that achieves 90% test accuracy on FashionMNIST dataset

Working with Modern Neural Network Framework - Classifying FashionMNIST

Designed a neural network using PyTorch that achieves 90% test (not training) accuracy on the FashionMNIST dataset.

• Some modifications were made on:
  • Architecture
  • Hyperparameters
  • Learning rate, activation function, etc
  • Epochs trained

Fashion MNIST Dataset:

• Set of 70k black and white images of clothes
  • 60k training instances
  • 10k test instances
• Labels - one-hot encodings of 10 classes
  • 0 T-shirt/top
  • 1 Trouser
  • 2 Pullover
  • 3 Dress
  • 4 Coat
  • 5 Sandal
  • 6 Shirt
  • 7 Sneaker
  • 8 Bag
  • 9 Ankle boot

The Neural Network fulfills the following:

• Classify images of clothes to assist in the return process at a shipping facility.
  • People will load clothing items onto a conveyer belt which will bring them to a camera.
  • The camera will capture the image as a 28x28 grayscale picture, and the neural network should return one of the 10 different classes as a result.
  • The item will be shipped to its corresponding department.
• Train the netwrok using a training set of 60,000 pictures of clothes.
• Test the network againts 10,000 pictures of clothes.
  • The goal here is to make the network classify images of clothes with at least 90% accuracy to ensure the sorting process is beneficial.

Stretch goals, for future improvements:

• Achieve higher accuracy.
• Design your own neural networks in NumPy! (or similar): If you thought that the forward and backward propagation algorithms were cool or you want to get a deeper understanding of algorithms yourself, you can try to code your own! This is certainly doable but might be slower.

For more details about the application and the code; checkout bellow:

Fashion MNIST Neural Network Code

Refrences

Dense neural network for MNIST

Dense neural netwrok for CIFAR10

Deep neural network for MNIST

Deep neural network for CIFAR10