Description

Note:
• In Homework #1 mini-project assignment, you will build and train your “Hello, world” machine learning programs of computer vision based on Fashion MNIST dataset (https://github.com/zalandoresearch/fashion-mnist). Fashion MNIST is intended as a drop-in replacement for the classic MNIST dataset.
• Although we’ve mainly introduced TensorFlow, you are free to choose any other API to build your model. Please make sure to add comments to your code whenever needed.
• Please submit your Jupyter notebook or python code to Canvas.

1. (50 points)
Please build and train a simple deep neural network (DNN) to classify the 10 image classes as shown above following the steps below:
(1) (5 points) Import Fashion MNIST dataset
(2) (5 points) Explore (and preprocess if needed) your data
(3) (10 points) Setup three layers: input layer, hidden layer with 128 units using ReLu, and output layer
(4) (10 points) Compile your model: choose your loss function, optimizer, and metrics
(5) (10 points) Train your model
(6) (10 points) Evaluate the performance and make prediction based on 10 randomly selected testing images
Page 1 of 2
2. (50 points)
Please build and train a convolutional neural network (CNN) to classify the 10 image classes as shown above, to obtain a higher performance than the DNN model, following the steps below:
(1) Import Fashion MNIST dataset
(2) Explore (and preprocess if needed) your data
(3) (10 points) Setup the necessary layers: convolutional layer, pooling layer, dense layer, output layer
(4) (10 points) Compile your model: choose your loss function, optimizer, and metrics
(5) (10 points) Train your model
(6) (10 points) Evaluate the performance and make prediction based on 10 randomly selected testing images
(7) (10 points) Applying error analysis for your model and provide suggestion to improve your model’s performance

Page 2 of 2