Description
- Consider a four layer fully-connected (include input layer) network with , neurons in four layers respectively. Input are fed into the first layer and represented by x, the loss is mean squared error E, and the activation function for each layer is sigmoid function . Let the label vector be t of size and let each layer output vector be and the input for each layer be , both of size
How many trainable parameters in this model (You don’t need to consider the bias term in this exercise)
- Fill out the code cells in hw_8_tf.ipynb to get yourself familiar with Tensorflow.
- Go to the datasets folder and run the script get_datasets.sh to download cifar-10 dataset (for mac you can simply type “sh ./get_datasets.sh” to run the script). Fill out the code cells hw_8.ipynb. Include the question answer in your homework document submission as well as in the Jupyter notebook.
