Week3_S1

The main reason activation functions like ReLU are used between layers in a neural network is to

• Reduce vanishing gradients

• Increase the number of layers

• Speed up backpropagation

• Introduce non-linearity

What does the ‘Flatten’ layer in Keras do?

• Declares the expected shape of input data

• Optimizes model weights

• Converts multi-dimensional data into a flat 1D vector

• Reduces overfitting

Which Keras API is most suited for building models by stacking layers sequentially?

• Functional API

• Gradient API

• Sequential API

• Model Subclassing

In Keras, the __________ layer is used to perform weighted sum and apply activation.

13. What happens if all network weights are initialized to zero?

• Faster convergence

• Different neurons will learn unique patterns

• Model overfits immediately

• All neurons in a layer will behave identically

In Keras, the __________ API is ideal for defining complex, non-linear model architectures.

32. When building a neural network for binary classification, the output layer typically has

• Two neurons with sigmoid activation

• One neuron with sigmoid activation

• Two neurons with softmax activation

• Three neurons with tanh activation

When a network is too deep without proper initialization or normalization, it can suffer from

• Vanishing gradients

• Linear explosion

• Exploding gradients

• Weight convergence

What happens if the learning rate is set too high?

• The model converges slowly but steadily

• Training becomes deterministic

• The loss oscillates and fails to converge

• The model always overfits

An epoch is defined as one complete pass through the entire __________ dataset.