Quiz Questions – Units 3, 4, and 5

What is the main feature that distinguishes RNNs from feedforward neural networks?

• Convolutional filters

• Feedback connections

• Weight sharing across layers

• Batch normalization

Unfolding an RNN in time creates a structure similar to:

• A CNN

• A feedforward network with shared weights

• A deep belief network

• A Boltzmann machine

Which RNN design pattern converts an input sequence into a single output value?

• Encoder

• Acceptor

• Transducer

• Decoder

The Encoder-Decoder architecture is mainly used in:

• Classification

• Regression

• Sequence-to-sequence tasks

• Feature extraction

What is the major problem when training RNNs over long sequences?

(a)  

In a Bidirectional RNN, the network processes the sequence:

• Only forward

• Only backward

• Both forward and backward

• Randomly

Sequence-to-sequence RNN models are widely used in:

• Object detection

• Machine translation

• Image segmentation

• Speech enhancement

Deep Recurrent Networks are formed by:

• Stacking multiple RNN layers

• Adding convolutional layers

• Using more neurons in one layer

• Reducing time steps

Recursive Neural Networks are useful for:

• Temporal sequence modeling

• Hierarchical data like parse trees

• Image classification

• Time series prediction

Which of the following helps to reduce vanishing gradient problems?

• Prevent overfitting

• Handle non-linearity

• Reduce vanishing gradient problems

• Increase network depth