Deep Learning Assessment

Which algorithms below use neural networks? (select all that apply)

• CNN

• K-means clustering

• Transformers

• Deep decision trees

What are the key components of a layer in a basic neural network? (Select all that apply)

• Weights and biases

• Activation functions

• Loss functions

• Learning rate

What is an epoch in the context of training neural networks?

• The number of layers in a neural network.

• The number of times the learning algorithm goes through the entire training dataset.

• The number of neurons in each layer of the network.

• The number of batches the data is divided into during training.

In a neural network for image recognition (e.g., identifying objects in photos), what is the role of different layers in the network? For example, how might the first few layers process a photo of a cat compared to the deeper layers?

• The first few layers recognize high-level features like the entire cat, while deeper layers focus on individual pixels.

• The first few layers detect simple features like edges and textures, while deeper layers recognize more complex patterns like shapes, eyes, or entire objects.

• All layers process the image in the same way, with each layer improving accuracy by repeating the same computation.

• The first few layers detect colors, while deeper layers detect objects based solely on size.

Which of the following statements about Artificial Neural Networks (ANNs) is/are correct?

• They are inspired by biological neurons and can learn complex non-linear relationships.

• They require less data and fewer computations than traditional machine learning algorithms.

• Their performance often improves with more layers and neurons, but risk overfitting.

• They cannot be used for regression tasks.

Which of the following about ANN and CNN is true? (select all that apply)

• ANNs are only used for regression tasks, while CNNs are used for classification.

• CNNs use convolutional layers to process data with a grid-like topology, such as images.

• ANNs use recurrent connections to handle sequential data, while CNNs do not.

• CNNs require less memory than standard ANNs for the same number of neurons

When choosing between a Convolutional Neural Network (CNN) and a traditional ANN for image-related tasks, which factor(s) primarily motivate using a CNN?

• CNNs require significantly less labeled data than ANNs.

• CNNs use spatial hierarchies, making them better at detecting local features in images.

• CNNs can handle sequential data better than ANNs.

• CNNs use shared weights (convolution filters) to reduce the number of parameters in comparison to fully connected layers.

Which of the following is true about Recurrent Neural Networks (RNNs)?**

• RNNs are well-suited for processing fixed-size input data.

• RNNs are specialized for image recognition.

• RNNs are ideal for handling sequential data, such as time series or text.

• RNNs do not use backward propagation

Which of the following are commonly cited advantages of using Recurrent Neural Networks (RNNs) for sequence data?

• They can handle variable-length inputs by using recurrent connections.

• They are immune to vanishing or exploding gradients.

• They allow parameters to be shared across different time steps.

• They can naturally process input sequences in one direction (forward) or both directions (bidirectional).

Question: Which of the following is true about dropout layer?

• dropout layer can help mitigate overfitting
  
  

• Using  dropout is a method to increase training speed.

• Applying dropout naively in recurrent connections can break the temporal dependencies

• Dropout is only beneficial in small datasets.