AI AI Engineer: Neural Networks and Deep Learning 2

Question 1

What is a recurrent neural network (RNN) primarily designed to handle?

Accepted Answer

Sequential and time-series data with temporal dependencies

Answer

Image classification

Answer

Tabular data regression

Answer

Clustering unlabeled data

Question 2

What problem do LSTM (Long Short-Term Memory) networks solve compared to vanilla RNNs?

Accepted Answer

Long-range dependency learning by mitigating vanishing gradients over long sequences

Answer

Slow matrix multiplication

Answer

Overfitting on small datasets

Answer

High memory consumption during inference

Question 3

In deep learning, what is transfer learning?

Accepted Answer

Reusing a model pretrained on a large dataset as a starting point for a new task

Answer

Moving a model from one server to another

Answer

Transferring data between training and validation sets

Answer

Sharing weights between two different models

Question 4

What is the role of the 'softmax' activation function in a multi-class classification output layer?

Accepted Answer

Converts raw logits into a probability distribution summing to 1

Answer

Introduces non-linearity in hidden layers

Answer

Normalizes input features

Answer

Reduces dimensionality of the output

Question 5

Which deep learning architecture introduced skip (residual) connections to enable training of very deep networks?

Accepted Answer

ResNet

Answer

VGG

Answer

AlexNet

Answer

GoogLeNet

Question 6

What is the purpose of weight initialization in neural networks?

Accepted Answer

To set initial parameter values that promote stable gradient flow at the start of training

Answer

To set hyperparameters before training

Answer

To define the network architecture

Answer

To normalize the training data