README.md

Fashion MNIST Classifier

Implementation

• Original model
• Underlying features

Scoring

With 1 dense layer

Overall accuracy: 0.7950

With 2 dense layers

Baseline accuracy: 0.8484
Best validation accuracy: 0.8640

How this neural network is structured?

Input, output and hidden layers

Summary

Model

Type: "sequential"

Layer (type)	Output Shape	Param #
flatten (Flatten)	(None, 784)	0
dense (Dense)	(None, 256)	200960
dense_1 (Dense)	(None, 128)	32896
dropout (Dropout)	(None, 128)	0
dense_2 (Dense)	(None, 10)	1290

Total params: 235,146
Trainable params: 235,146
Non-trainable params: 0

Optimizer

Adam (SGD)
Adam optimization is a stochastic gradient descent method that is based on adaptive estimation of first-order and second-order moments.

Loss Function

Sparse categorical crossentropy

References

Xiao, H. et al. Fashion MNIST Dataset. GitHub; 2015.
Sampaio, C. Deep Learning. AOVS Sistemas de Informática S.A; 2019.
Chollet, F. et al. Keras. GitHub; 2015.