Artificial Intelligence Artificial Intelligence 3 — Questions and Answers
Question 1: Which technique helps prevent overfitting by randomly disabling neurons during training?
- Batch normalization
- Gradient clipping
- Dropout (Correct answer)
- Momentum
Correct answer: Dropout
Dropout randomly zeroes neuron outputs during training, forcing the network to learn redundant representations.
Question 2: In natural language processing, what does tokenization do?
- Splits text into smaller units like words or subwords (Correct answer)
- Translates text between languages
- Removes grammatical errors
- Encrypts sensitive text
Correct answer: Splits text into smaller units like words or subwords
Tokenization breaks raw text into discrete units that models can process numerically.
Question 3: A convolutional neural network (CNN) is most commonly associated with which type of data?
- Tabular spreadsheets
- Images (Correct answer)
- Audio transcripts only
- Graph databases
Correct answer: Images
CNNs exploit spatial locality with convolutional filters, making them ideal for image data.
Question 4: What is the primary purpose of the minimax algorithm in game-playing AI?
- To maximize training data usage
- To reduce memory consumption
- To cluster game states
- To choose moves assuming an optimal opponent (Correct answer)
Correct answer: To choose moves assuming an optimal opponent
Minimax selects moves that maximize the player's worst-case outcome against a perfectly playing adversary.
Question 5: Which scenario best illustrates algorithmic bias in AI?
- A model running slower on older hardware
- A chatbot going offline during maintenance
- A hiring model penalizing resumes from certain demographics due to skewed training data (Correct answer)
- A model requiring GPU acceleration
Correct answer: A hiring model penalizing resumes from certain demographics due to skewed training data
Bias arises when skewed training data causes systematically unfair outcomes for particular groups.
Question 6: What does backpropagation compute during neural network training?
- Gradients of the loss with respect to each weight (Correct answer)
- The optimal network architecture
- The number of required epochs
- The best learning rate
Correct answer: Gradients of the loss with respect to each weight
Backpropagation applies the chain rule to compute how each weight contributes to the loss, enabling gradient descent updates.
Question 7: Which of the following describes narrow AI?
- AI that can perform any intellectual task a human can
- AI with self-awareness
- AI that requires no training data
- AI specialized in a single task like playing chess (Correct answer)
Correct answer: AI specialized in a single task like playing chess
Narrow (weak) AI excels at one specific task but cannot generalize beyond its designed domain.
Artificial Intelligence Artificial Intelligence 3
Which technique helps prevent overfitting by randomly disabling neurons during training?