PSI: 9 Deep Learning (2/2)

A CNN can be viewed as a set of

• Spatial filters
• Autoencoders
• Recurrent layers
• All the others

The pooling layer of a CNN is used for

• Avoiding the vanishing gradient problem
• Performing a convolution
• Dimensionality reduction
• None of the others

If you want to manage input patterns that change with time you should use

• Convolutional NN
• Recurrent NN
• Self-organizing Maps
• Multilayer Perceptron

When your model performs badly with the training data you are suffering

• Underfitting
• Overfitting
• Cross-fitting
• Cross-validation

A solution for reducing the time to train a RNN is

• Praying
• Using a ReLU
• Gating
• Pooling