Artificial Intelligence Artificial Intelligence 5 — Questions and Answers
Question 1: Which technique adapts a large pre-trained model to a specific task using a smaller labeled dataset?
- Random initialization
- Fine-tuning (Correct answer)
- Data augmentation
- Feature hashing
Correct answer: Fine-tuning
Fine-tuning continues training a pre-trained model on task-specific data, leveraging transfer learning.
Question 2: An AI chatbot confidently states a false fact that sounds plausible. What is this phenomenon called?
- Regularization
- Quantization
- Tokenization
- Hallucination (Correct answer)
Correct answer: Hallucination
Hallucination refers to language models generating fluent but factually incorrect or fabricated content.
Question 3: In a Markov decision process, what property must state transitions satisfy?
- They depend only on the current state and action (Correct answer)
- They depend on the entire history of states
- They must be deterministic
- They must have equal probabilities
Correct answer: They depend only on the current state and action
The Markov property requires that the next state depends only on the current state and action, not past history.
Question 4: Which evaluation approach splits data into multiple folds, training and testing on different combinations?
- Grid search
- Early stopping
- Cross-validation (Correct answer)
- Bootstrapping weights
Correct answer: Cross-validation
K-fold cross-validation rotates which fold is held out for testing, giving a more reliable performance estimate.
Question 5: A recommendation system suggests movies based on what similar users enjoyed. What is this approach called?
- Content-based filtering
- Rule-based inference
- Genetic programming
- Collaborative filtering (Correct answer)
Correct answer: Collaborative filtering
Collaborative filtering recommends items using the preferences of users with similar tastes.
Question 6: What is the main advantage of using word embeddings over one-hot encoding in NLP?
- They capture semantic similarity between words in dense vectors (Correct answer)
- They require no training data
- They eliminate the need for tokenization
- They guarantee perfect translation
Correct answer: They capture semantic similarity between words in dense vectors
Embeddings place semantically similar words close together in vector space, unlike sparse one-hot vectors.
Question 7: Which concern is central to explainable AI (XAI)?
- Reducing electricity costs
- Making model decisions understandable to humans (Correct answer)
- Increasing model size
- Compressing training datasets
Correct answer: Making model decisions understandable to humans
XAI aims to make black-box model decisions transparent and interpretable, which is critical in high-stakes domains.
Artificial Intelligence Artificial Intelligence 5
Which technique adapts a large pre-trained model to a specific task using a smaller labeled dataset?