advance ai midterm study

Question 1

Which is a problem with character tokenization?

Accepted Answer

Characters don’t carry semantic meaning like words

Question 2

Which of these is NOT a common performance metric in machine learning?

Accepted Answer

ADA

Question 3

What makes Convolutional Autoencoders (CAEs) different from standard autoencoders?

Accepted Answer

They use convolutional layers instead of fully connected layers

Question 4

What challenge does scalability address in ML systems?

Accepted Answer

Handling increasing amounts of data and computation efficiently

Question 5

Transformers are used only for language-related tasks

Accepted Answer

False

Question 6

What comes immediately after 'Model Training' in the ML pipeline?

Accepted Answer

Evaluation

Question 7

What are the two main Word2Vec architectures?

Accepted Answer

Continuous Bag-of-Words (CBOW) and Continuous Skip-gram

Question 8

What problem does Word2Vec help solve in natural language processing?

Accepted Answer

Encoding words into dense vectors that capture meaning

Question 9

Transformers process entire sequences in parallel instead of step by step like RNNs.

Accepted Answer

True

Question 10

Which of the following is true about Encoder–Decoder networks?

Accepted Answer

They enable end-to-end training for translation tasks

Question 11

What is a key limitation of using fully connected layers for image input?

Accepted Answer

They treat each pixel independently

Question 12

What is the main purpose of an Encoder–Decoder network in neural machine translation?

Accepted Answer

convert a input sentence into a fixed-size context vector

Question 13

Why is the vanishing gradient problem bad for learning long-range dependencies?

Accepted Answer

Small gradients make updates too tiny to learn distant patterns

Question 14

What does Byte Pair Encoding (BPE) do?

Accepted Answer

Splits text into frequent character or sub-word pairs to handle rare words

Question 15

This auto encoder diagram increases from 2 data values to 4 data values and is a good example of a auto encoder

Accepted Answer

False

Question 16

Autoencoders are especially useful for which of the following tasks?

Accepted Answer

Dimensionality reduction

Question 17

Word tokenization treats every unique word as a separate token, which can lead to a very large vocabulary

Accepted Answer

True

Question 18

What is the main function of an autoencoder?

Accepted Answer

Compress data into a smaller representation

Question 19

Why does the basic encoder–decoder model struggle with long sentences?

Accepted Answer

It compresses the entire input into one fixed-size vector

Question 20

What happens to embedding vectors during model training?

Accepted Answer

Adjusted along with task-specific model weights to improve performance

Question 21

What is a key advantage of GRUs compared to LSTMs?

Accepted Answer

They are simpler and more efficient

Question 22

How does the decoder in a CAE typically reconstruct the image?

Accepted Answer

Using transposed convolutions

Question 23

Which of the following is NOT a typical step in the ML pipeline?

Accepted Answer

Web Development

Question 24

What is a Long Short-Term Memory (LSTM) network designed to do?

Accepted Answer

Model sequences and remember long-term dependencies

Question 25

What key mechanism do Transformers use to understand relationships in sequences?

Accepted Answer

Self-attention

Question 26

What is the "vanishing gradient problem" in neural networks?

Accepted Answer

When gradients become too small during training

Question 27

What is the role of the embedding layer in an NLP model?

Accepted Answer

To convert token IDs into dense vectors representing word meaning

Question 28

Why do traditional feedforward neural networks struggle with sequences like sentences

Accepted Answer

They assume all inputs are independent

Question 29

The decoder generates the translation word by word using the encoder’s context vector

Accepted Answer

True

Question 30

Transformers understand word order naturally because they have recurrence like RNN

Accepted Answer

False

Question 31

Why is positional encoding used in Transformer models?

Accepted Answer

To help the model understand the order of words

Question 32

Word embeddings capture semantic meaning based on the context words appear in.

Accepted Answer

True

Question 33

Word embedding represents words as dense vectors that capture their meaning based on context.

Accepted Answer

True

Question 34

Which type of neural network is especially affected by the vanishing gradient problem?

Accepted Answer

Recurrent Neural Networks (RNNs)

Question 35

GRUs are more complex alternative to LSTMs for handling sequential data

Accepted Answer

False

Question 36

What’s a key advantage of Recurrent Neural Networks (RNNs)?

Accepted Answer

They can remember previous inputs in a sequence

Question 37

Which of the following is an example of word tokenization?

Accepted Answer

"I love pizza" → ["I", "love", "pizza"]

Question 38

What happens when we flatten a 2D image into a 1D vector?

Accepted Answer

It removes spatial relationships between pixels

Question 39

LSTMs solve the vanishing gradient problem using memory cells and gates.

Accepted Answer

True

Question 40

Embeddings are randomly assigned to words and do not change during training

Accepted Answer

False

Question 41

Word2Vec represents words as sparse one-hot vectors.

Accepted Answer

False

Question 42

What is sub-word tokenization?

Accepted Answer

Splitting words into smaller meaningful pieces between words and characters

advance ai midterm study

Questions 42

advance ai midterm study

Questions 42