Practice for the AI-900: Microsoft Azure AI Fundamentals exam with 13 exam-style practice questions, instant answer reveals, and concise explanations of every correct answer. Topics include: A company uses Azure AI services for speech-to-text, text translation, and sentiment analysis. They want to simplify bil. Follow @CertPunch and visit certpunch.com for more certification practice exams and study content.
Prefer hands-on? Take this round as an interactive practice test — answer every question, get instant feedback, and see your score: Start the AI-900: Microsoft Azure AI Fundamentals practice test →
What you will practice
- A company uses Azure AI services for speech-to-text, text translation, and sentiment analysis. They want to s…
- Which statement BEST describes the relationship between model complexity and responsible AI implementation?
- A company wants to train a model to recognize rare bird species in forest images. The prebuilt Azure AI Visio…
- You are designing a prompt for a large language model to summarize a technical article. Which element of the…
- In evaluating a classification model, which metric is most critical when false positives are particularly cos…
- A retail app uses Azure AI Vision to generate descriptive text for product images, such as "a red sweater on…
Answers and explanations
Tap a question to expand the answer and the exam reasoning. Try to commit to your own pick first.
Q1. A company uses Azure AI services for speech-to-text, text translation, and sentiment analysis. They want to simplify billing and management. Which type of resource should they create in Azure?
Answer: B. A multi-service resource
A multi-service resource lets you manage multiple Azure AI services with a single endpoint and key. This consolidation directly simplifies billing and administration. Single-service resources require separate management, increasing overhead.
Q2. Which statement BEST describes the relationship between model complexity and responsible AI implementation?
Answer: D. Model complexity should be balanced against interpretability based on use case requirements
Responsible AI requires balancing model complexity with interpretability based on specific use case requirements. While complex models offer high accuracy, they often act as black boxes, meaning simpler models are sometimes preferred to maintain necessary transparency.
Q3. A company wants to train a model to recognize rare bird species in forest images. The prebuilt Azure AI Vision models lack these species. Which approach should they use?
Answer: B. Use Azure Custom Vision with labeled bird images
Azure Custom Vision allows you to train custom image classification models using your own labeled datasets. This service is specifically designed to extend prebuilt capabilities when you need to recognize highly specialized or niche visual subjects.
Q4. You are designing a prompt for a large language model to summarize a technical article. Which element of the prompt explicitly defines the task for the model?
Answer: D. Instructions
Instructions explicitly tell the model exactly what task to perform, such as summarizing the provided text. Context, input data, and output indicators simply provide supporting details and formatting cues rather than the core command.
Q5. In evaluating a classification model, which metric is most critical when false positives are particularly costly?
Answer: C. Precision
Precision measures the proportion of true positives among all positive predictions, making it crucial when false positives are costly. Recall is the opposite, prioritizing the identification of all actual positives regardless of false alarms.
Q6. A retail app uses Azure AI Vision to generate descriptive text for product images, such as "a red sweater on a mannequin." Which capability combines computer vision and natural language processing (NLP) to achieve this?
Answer: B. Dense Captions
Dense Captions combines computer vision and natural language processing to generate detailed descriptive text for images. Optical character recognition only extracts written text, rather than describing the actual visual contents of the provided image.
Q7. A company wants to integrate a pre-trained Hugging Face model into their Azure AI solution. Which feature of Azure AI Foundry should they use?
Answer: B. Model Catalog
The Model Catalog in Azure AI Foundry provides a central repository of pre-trained models, including those from Hugging Face. This feature allows developers to easily browse and deploy models without needing to build them entirely from scratch.
Q8. Why is an original dataset typically split into training, validation, and test sets?
Answer: B. To allocate distinct data for learning the patterns, tuning hyperparameters, and obtaining an unbiased evaluation on unseen examples.
Splitting data allocates distinct sets for learning, tuning, and final evaluation to ensure generalization. If you test on training data, you risk overfitting, which means the model simply memorizes examples rather than learning actual patterns.
Q9. A model achieves 98% accuracy on training data but only 60% on validation data. During training, the model's error decreases steadily, but validation error starts rising after 50 epochs. What is the MOST likely issue, and how can it be add…
Answer: B. Overfitting – Introduce dropout layers or early stopping.
Rising validation error while training error decreases indicates the model is overfitting to the training data. Techniques like dropout layers or early stopping prevent the model from memorizing noise and help it generalize better to new data.
Q10. A developer is experimenting with a large language model that uses tokenization as a first step in processing text. Which statement best describes the role and nature of tokens in these models?
Answer: B. Tokens are the basic units of text and can represent whole words, subwords, or even characters, depending on the tokenizer design.
Tokens are the basic units of text and can represent whole words or subwords depending on the tokenizer design. They are not fixed arrays; this chunking process helps language models handle diverse inputs and efficiently process natural language.
Q11. A hospital uses a model to detect a rare disease. The priority is to ensure no affected patients are missed, even if some healthy patients are incorrectly flagged. Which metric should be prioritized to evaluate this model?
Answer: B. Recall
Recall measures the proportion of actual positive cases correctly identified, minimizing false negatives. In medical scenarios where missing a disease is dangerous, prioritize recall over precision, because precision mainly focuses on minimizing false positives.
Q12. During image processing, a small matrix (kernel) can be systematically applied across an image. What is the primary function of this kernel when performing a convolution on an image?
Answer: A. To detect features by emphasizing areas with significant changes in pixel intensity.
The kernel acts as a filter designed to highlight specific features, such as edges, by computing a weighted sum of pixel intensities to emphasize regions with sharp changes. For the AI-900 exam, associate convolutions and kernels directly with feature detection in computer vision.
Q13. Regarding Azure's face recognition services, which statement correctly differentiates between face verification and face identification?
Answer: A. Face verification performs a one-to-one comparison to confirm an individual's identity, whereas face identification conducts a one-to-many search within a repository to find matching identities.
Face verification performs a one-to-one comparison to confirm an individual's identity against a saved image. Face identification conducts a one-to-many search within a repository to find matching identities. Remember that verification equals one-to-one, while identification equals one-to-many.
More AI-900: Microsoft Azure AI Fundamentals drills and other practice exams are on @CertPunch. New rounds drop every few days at certpunch.com.