Microsoft Azure AI Fundamentals Microsoft Azure AI Fundamentals 4 — Questions and Answers
Question 1: What is the role of a 'validation dataset' during model training?
- To provide the initial data for model training
- To tune hyperparameters and monitor for overfitting during training (Correct answer)
- To give the final unbiased evaluation of model performance
- To label raw unlabeled data before training
Correct answer: To tune hyperparameters and monitor for overfitting during training
The validation dataset is used during training to tune hyperparameters and detect overfitting, while remaining separate from the final test set.
Question 2: Azure Personalizer uses which type of machine learning to deliver personalized content recommendations?
- Supervised learning
- Unsupervised learning
- Reinforcement learning (Correct answer)
- Transfer learning
Correct answer: Reinforcement learning
Azure Personalizer uses reinforcement learning, where it learns which actions (content selections) yield the best rewards (user engagement) over time.
Question 3: Which feature of Azure Computer Vision can describe the content of an image in natural language?
- OCR (Optical Character Recognition)
- Image captioning / Describe Image (Correct answer)
- Thumbnail generation
- Color scheme analysis
Correct answer: Image captioning / Describe Image
The 'Describe Image' feature in Azure Computer Vision generates a human-readable sentence describing the content and context of an image.
Question 4: What is the key difference between 'precision' and 'recall' in a classification model?
- Precision measures training speed; recall measures inference speed
- Precision measures how many predicted positives are correct; recall measures how many actual positives were found (Correct answer)
- Precision applies to regression; recall applies to classification
- Precision is for binary models; recall is for multi-class models
Correct answer: Precision measures how many predicted positives are correct; recall measures how many actual positives were found
Precision = true positives / (true positives + false positives); Recall = true positives / (true positives + false negatives) — they measure different error trade-offs.
Question 5: Which Azure service allows you to build a question-answering system from existing documents and FAQs without training a custom model from scratch?
- Azure Custom Vision
- Azure Language Service (Question Answering) (Correct answer)
- Azure Anomaly Detector
- Azure Form Recognizer
Correct answer: Azure Language Service (Question Answering)
The Question Answering feature in Azure Language Service (formerly QnA Maker) extracts Q&A pairs from documents and FAQs to create a searchable knowledge base.
Question 6: In the context of natural language processing, what is 'tokenization'?
- Assigning sentiment scores to sentences
- Breaking text into individual words or subword units for processing (Correct answer)
- Translating text from one language to another
- Identifying named entities like people and organizations
Correct answer: Breaking text into individual words or subword units for processing
Tokenization splits raw text into smaller units (tokens) such as words or subwords, which are the fundamental inputs that NLP models process.
Question 7: What does 'confidence score' represent when returned by an Azure Cognitive Services prediction?
- The number of training samples used
- The model's estimated probability that its prediction is correct (Correct answer)
- The latency of the API response in milliseconds
- The version number of the deployed model
Correct answer: The model's estimated probability that its prediction is correct
A confidence score (typically 0–1 or 0–100%) indicates how certain the model is about its prediction — higher scores mean the model is more confident.
Microsoft Azure AI Fundamentals Microsoft Azure AI Fundamentals 4
What is the role of a 'validation dataset' during model training?