Deep Learning Recurrent Neural Networks 2

Question 1

What is a bidirectional RNN?

Accepted Answer

An RNN that processes the sequence in both forward and backward directions simultaneously

Answer

An RNN that processes images from multiple angles

Answer

An RNN with two hidden layers stacked vertically

Answer

An RNN that uses two different loss functions

Question 2

In sequence-to-sequence (Seq2Seq) models, what is the role of the encoder?

Accepted Answer

Compressing the input sequence into a fixed-length context vector

Answer

Generating the output sequence token by token

Answer

Applying attention over the input

Answer

Computing cross-entropy loss over predictions

Question 3

What is teacher forcing in RNN training?

Accepted Answer

Feeding the ground-truth previous output as the next input during training rather than the model's own prediction

Answer

Using a separate teacher network to guide training

Answer

Clipping gradients during backpropagation

Answer

Initializing weights from a pretrained language model

Question 4

What is gradient clipping used for when training RNNs?

Accepted Answer

Capping gradient magnitudes to prevent exploding gradients from destabilizing training

Answer

Preventing the loss from going below zero

Answer

Increasing the learning rate dynamically

Answer

Removing neurons with zero gradients

Question 5

What type of tasks are RNNs / LSTMs most naturally suited for?

Accepted Answer

Sequential and time-series tasks such as language modeling, speech recognition, and forecasting

Answer

Static image classification

Answer

Graph-structured data

Answer

Clustering unlabeled datasets

Question 6

What is the hidden state in an RNN?

Accepted Answer

A vector summarizing information from previous time steps passed to the next step

Answer

The final layer's output probabilities

Answer

The loss value at each step

Answer

The weight matrix connecting input to hidden layer