AI-900 Practice Exam 57 — 20 Questions

Question 1

What does the 'image analysis' API in Azure AI Vision return when given an image?

Accepted Answer

Rich metadata including captions, detected objects, tags, colour analysis, and content flags. The Image Analysis API in Azure AI Vision returns rich metadata about the image content, including captions, detected objects, tags, color analysis, and content moderation flags. This is because the API applies pre-trained deep learning models to extract semantic information from the image, not raw pixel data or aesthetic scores.

Answer

The raw pixel data of the image in a compressed format

Answer

A score from 1 to 10 rating the aesthetic quality of the photograph

Answer

A list of similar images found across the web

Question 2

What is 'word sense disambiguation' (WSD) and why is it challenging for NLP?

Accepted Answer

Determining which meaning of an ambiguous word is intended based on surrounding context. Word sense disambiguation (WSD) is the NLP task of identifying which specific meaning of a polysemous word (a word with multiple meanings) is intended in a given context, using surrounding words, syntax, and semantic cues. This is challenging because many words have multiple, often unrelated meanings (e.g., 'bank' as a financial institution vs. river bank), and the correct sense depends on subtle contextual signals that are difficult for models to capture without deep understanding of the domain or world knowledge.

Answer

Correcting spelling mistakes caused by homophones (words that sound the same)

Answer

Translating words from one language to their exact equivalent in another language

Answer

Measuring how many distinct meanings a word has across a dictionary

Question 3

What is 'key phrase extraction' in Azure AI Language?

Accepted Answer

Identifying the most important words and phrases that best represent a text's main topics. Key phrase extraction in Azure AI Language uses natural language processing to identify the most salient words and phrases that summarize the main topics of a text. It analyzes the document's structure and semantics to return a ranked list of key phrases, enabling quick understanding of core content without reading the entire text.

Answer

Encrypting sensitive phrases in a document for secure storage

Answer

Finding and extracting password-like phrases from user messages for security monitoring

Answer

Selecting the highest-scoring responses from a list of candidate answers

Question 4

What is 'image segmentation' and how does it differ from object detection?

Accepted Answer

Classifying every pixel in an image to identify precise boundaries — more detailed than bounding-box object detection. Image segmentation classifies every pixel in an image into a category, producing pixel-level masks that outline objects with precise boundaries. This differs from object detection, which only draws bounding boxes around objects and does not distinguish object edges or overlapping instances. Option B correctly captures this higher granularity and accuracy.

Answer

Dividing an image file into smaller files for distributed storage

Answer

Removing the background from an image by detecting edges

Answer

Dividing the training dataset into segments for cross-validation

Question 5

A game development studio uses Azure OpenAI Service to generate unique backstories for non-player characters (NPCs). They want the generated stories to be coherent and relevant to a given character class (e.g., warrior, mage) but also creative and varied. Which parameter should the studio adjust primarily to increase the creativity and variety of the generated text?

Accepted Answer

Increase the temperature parameter. Increasing the temperature parameter makes the model's output more random by scaling the probability distribution over tokens, which encourages less likely word choices and thus increases creativity and variety in generated text. For the game studio, a higher temperature (e.g., 0.8–1.0) will produce more diverse and imaginative backstories for different character classes, while still maintaining coherence if not set too high.

Answer

Increase the top_p parameter

Answer

Increase the frequency_penalty parameter

Answer

Decrease the max_tokens parameter

Question 6

A company develops an AI-powered virtual assistant for customer service. To ensure the assistant can be used by people with visual impairments, the team integrates screen reader compatibility. Which Microsoft responsible AI principle is most directly addressed by this action?

Accepted Answer

Inclusiveness. Integrating screen reader compatibility directly addresses the inclusiveness principle of responsible AI. This principle ensures that AI systems are designed to be accessible and usable by people with diverse abilities, including those with visual impairments, by supporting assistive technologies like screen readers.

Answer

Fairness

Answer

Reliability & Safety

Answer

Privacy & Security

Question 7

What is 'few-shot learning' in the context of Azure AI Custom Vision model training?

Accepted Answer

Training an accurate vision model with very few labelled examples using transfer learning. Few-shot learning in Azure AI Custom Vision refers to training an accurate vision model with very few labeled examples by leveraging transfer learning. This approach uses a pre-trained neural network (e.g., ResNet) as a starting point, allowing the model to learn new visual concepts from as few as 2–5 images per class, significantly reducing the data collection burden.

Answer

Training a model using only a small subset of available compute resources

Answer

A technique for running multiple small training experiments in parallel

Answer

Limiting training to the first few hundred iterations regardless of convergence

Question 8

What is 'speaker diarisation' in Azure AI Speech and when is it used?

Accepted Answer

Identifying and labelling which speaker said which portions of a multi-speaker audio recording. Speaker diarization is an Azure AI Speech feature that segments an audio recording by speaker identity, labeling each segment with a unique speaker tag (e.g., Speaker 1, Speaker 2). It is used in scenarios like meeting transcription, call center analytics, or any multi-speaker audio where distinguishing who spoke when is required. This directly matches option B's description of identifying and labeling which speaker said which portions of a multi-speaker recording.

Answer

Translating spoken audio into the dialect of the speaker's home region

Answer

Detecting when a speaker is lying based on vocal stress patterns

Answer

Counting how many unique speakers have interacted with an AI voice assistant over time

Question 9

What is 'span-based named entity recognition' and how does it differ from sequence labelling?

Accepted Answer

Identifying text spans as candidate entities and classifying each span — handling overlapping and nested entities. Span-based named entity recognition (NER) identifies candidate entities by first locating text spans (contiguous sequences of tokens) and then classifying each span into an entity type. This differs from sequence labeling (e.g., BIO tagging) because it can naturally handle overlapping and nested entities—for example, recognizing both "University of Washington" as an organization and "Washington" as a location within the same sentence—whereas sequence labeling typically assigns a single label per token and struggles with such hierarchies.

Answer

NER that works across spans (paragraphs) of text rather than single sentences

Answer

A technique that spans multiple languages to recognise entities in multilingual text

Answer

NER that spans multiple documents to track entities across a corpus

Question 10

What is 'multi-agent systems' in the context of Azure AI and agentic workflows?

Accepted Answer

Multiple specialised AI agents that collaborate — each with different roles — to accomplish complex goals. In Azure AI and agentic workflows, a multi-agent system involves multiple specialized AI agents, each with distinct roles (e.g., planner, coder, reviewer), that collaborate to decompose and solve complex tasks. This architecture leverages the Azure AI Agent Service to orchestrate agent communication and task delegation, enabling more robust and scalable solutions than a single monolithic model.

Answer

Running multiple instances of the same model simultaneously for load balancing

Answer

AI systems deployed across multiple Azure regions for global availability

Answer

Security agents that monitor AI systems for prompt injection and misuse

Question 11

What is 'natural language generation' (NLG) and how does it differ from NLU?

Accepted Answer

NLU is understanding language input; NLG is producing natural language output from data. Natural Language Generation (NLG) is the AI capability that produces coherent, human-readable text or speech from structured data or other inputs. It differs from Natural Language Understanding (NLU), which focuses on interpreting and extracting meaning from language input. Option B correctly identifies NLU as understanding input and NLG as generating output, which is the fundamental distinction between these two subfields of natural language processing (NLP).

Answer

NLG is the same as NLU — both involve processing natural language

Answer

NLG is a hardware component that accelerates language model inference

Answer

NLU works on text; NLG works only on spoken audio

Question 12

What is 'image embedding' in computer vision and how is it used in visual search?

Accepted Answer

Converting images to vectors that capture visual meaning for similarity search and retrieval. Image embedding converts images into dense vector representations (embeddings) that capture semantic visual features such as shapes, colors, and textures. In visual search, these embeddings enable similarity comparisons by calculating distances (e.g., cosine similarity) between query image vectors and a pre-indexed database of image vectors, allowing retrieval of visually similar images even without textual metadata.

Answer

Inserting an image into a Word document or web page as an embedded object

Answer

Compressing images before embedding them in a database to reduce storage costs

Answer

Annotating images with GPS coordinates embedded in the file metadata

Question 13

A data scientist trains a deep neural network on a small dataset. The model achieves 100% accuracy on the training data but only 60% accuracy on a validation set. Which technique is most appropriate to address this issue?

Accepted Answer

Apply regularization. The model's perfect training accuracy (100%) paired with poor validation accuracy (60%) is a classic sign of overfitting, where the model has memorized the training data rather than learning generalizable patterns. Regularization techniques (e.g., L1/L2 regularization, dropout) penalize large weights or randomly drop neurons during training, which forces the network to learn simpler, more robust features and reduces overfitting on small datasets.

Answer

Increase the number of training epochs

Answer

Add more hidden layers

Answer

Increase the learning rate

Question 14

What is the purpose of a test dataset in machine learning model development?

Accepted Answer

To provide an unbiased final evaluation of the trained model on unseen data. The test dataset is used to provide an unbiased final evaluation of the trained model on unseen data. This is critical in machine learning because the model has never seen the test examples during training or validation, so the evaluation metrics (e.g., accuracy, precision, recall) reflect the model's true generalization ability. In Azure Machine Learning, the test dataset is typically split from the original data before any training begins and is only used once at the end of the model development lifecycle.

Answer

To provide additional examples for training the model

Answer

To tune hyperparameters and select the best model version

Answer

To monitor model performance after deployment

Question 15

What is 'dimensionality reduction' and why is it useful in machine learning?

Accepted Answer

Reducing the number of input features while preserving key information for efficient modelling. Dimensionality reduction is the process of reducing the number of input features (variables) in a dataset while retaining as much of the original information as possible. This is useful in machine learning because it helps combat the 'curse of dimensionality', reduces overfitting, lowers computational cost, and can improve model performance by eliminating noise and redundant features. In Azure Machine Learning, techniques like Principal Component Analysis (PCA) are commonly used for this purpose.

Answer

Reducing the physical size of AI hardware components for edge deployment

Answer

Reducing the model's output to a single dimension for binary decision making

Answer

Simplifying the Azure ML workspace to have fewer compute resources and experiments

Question 16

What is 'model export' in Azure Custom Vision and what formats are supported?

Accepted Answer

Exporting trained models as ONNX, TensorFlow, CoreML, or Docker for offline/edge deployment. Model export in Azure Custom Vision allows you to export a trained image classification or object detection model in formats like ONNX, TensorFlow, CoreML, or Docker container images. This enables the model to run offline on edge devices or local servers without requiring a continuous connection to the Azure cloud, which is critical for low-latency or disconnected scenarios.

Answer

Exporting model training logs and metrics to Excel for analysis

Answer

Exporting the training data to another Azure service for fine-tuning

Answer

Exporting a Custom Vision project as a YAML configuration file for source control

Question 17

A company develops an AI system to screen job applications. The system is intended to be used by candidates who may have visual, hearing, or motor impairments. The company wants to ensure that the interface is accessible to all candidates regardless of disability. Which Microsoft responsible AI principle should they prioritize?

Accepted Answer

Inclusiveness. The scenario focuses on ensuring the interface is usable by candidates with visual, hearing, or motor impairments. Microsoft's Inclusiveness principle is specifically designed to address accessibility and ensure AI systems empower everyone, including people with disabilities, by designing for a wide range of human abilities. This principle directly guides the development of accessible interfaces, such as supporting screen readers, alternative input methods, and captioning.

Answer

Fairness

Answer

Reliability and safety

Answer

Transparency

Question 18

What is the Azure AI Language Studio used for in addition to testing built-in features?

Accepted Answer

Building and training custom NLP models including custom classification and NER. Azure AI Language Studio is a comprehensive tool that allows users to not only test pre-built language features but also to build, train, and deploy custom NLP models, such as custom text classification and custom named entity recognition (NER). This extends beyond simple testing to enable tailored solutions for domain-specific language understanding.

Answer

Only for testing pre-built language features without any customization

Answer

Writing and executing Python code for NLP tasks

Answer

Managing billing and API keys for Azure AI Language

Question 19

A news agency wants to build a system that can automatically generate a short headline for each news article. The system should select the most important sentence from the article as the headline. Which Azure AI Language feature should they use?

Accepted Answer

B: Extractive summarization. Extractive summarization is the correct Azure AI Language feature because it identifies and extracts the most important sentences from a document, which directly matches the requirement to select the most important sentence as a headline. Unlike key phrase extraction, which returns individual words or short phrases, extractive summarization returns complete sentences ranked by relevance, making it ideal for headline generation.

Answer

A: Key phrase extraction

Answer

C: Entity recognition

Answer

D: Sentiment analysis

Question 20

A company uses Azure OpenAI Service to generate summaries of long technical documents. They notice that the model sometimes produces summaries that sound plausible but contain factual errors contradicting the source document. Which concept describes this type of error in large language models?

Accepted Answer

Hallucination. Hallucination in large language models refers to the generation of content that is factually incorrect or nonsensical but presented with confidence. In this scenario, the model produces summaries that sound plausible yet contain factual errors contradicting the source document, which is the hallmark of hallucination. This occurs because the model generates text based on probabilistic patterns rather than verifying facts against the input.

Answer

Overfitting

Answer

Tokenization

Answer

Bias