See more

… into a 30-dimensional vector, which is fed into a sequence of fully connected Dense hidden layers (700k parameters) with non-linear activations (relu). This produces a final vector of length 10 which captures all o…

Dense

relu

Using the Embedding layer in Keras, we embed the 🍞 into a 10-dimensional vector space. This uses 10 million parameters since we have 1 million potential products to consider. Similarly, …

Embedding