A neural network and deep learning project to create a custom model that follows the given requirements. The model consists of a Stem which takes an image and divides it into 4 patches of 14x14. Each patch is then vectorized and transformed to the feature vector. Next a linear layer is used which then is passed to the backbone. The Backbone consists of 2 blocks which then consists of 2 multi-layer perceptron (MLP). Each MLP first has a non-activation function, in this case the ReLU function is used. Following that, LeakyReLU function is used. The output is then passed onto the Classifier. The final model accuracy is 88.09%.

• Source CodeFashion-MNIST-Classification
• PlatformWindows/MacOS/Linux
• StackPython, Google Colab, Jupyter Notebook