Microsoft Updated AI-900 Exam Questions and Answers by aniyah

According to the Microsoft Azure AI Fundamentals (AI-900) official study materials and Microsoft Learn module “Describe features of common AI workloads,” an anomaly detection workload is designed to identify data points or patterns that deviate significantly from what is expected or normal. These anomalies often indicate irregularities, faults, or potential issues that require attention.

In this scenario, the AI system monitors temperature data from a large machine. Normally, the machine operates within a predictable temperature range. When the AI detects sudden or unexpected temperature spikes or drops — behavior that does not match the historical pattern — it flags these occurrences as anomalies. This type of workload is fundamental in predictive maintenance and industrial monitoring, where it helps detect equipment failures, safety hazards, or energy inefficiencies before they escalate.

Microsoft’s AI-900 curriculum emphasizes that anomaly detection workloads are often used in:

Industrial IoT systems (detecting abnormal sensor readings or machine behavior)

Finance (fraud detection or unusual transaction monitoring)

Cybersecurity (detecting irregular network traffic or access patterns)

Operations (identifying abnormal variations in production data)

The Azure service used for this purpose is Azure Anomaly Detector, part of Azure Cognitive Services, which uses advanced statistical and machine learning models to automatically detect outliers in time-series data such as temperature, pressure, or transaction logs.

By comparison:

Computer vision handles image or video analysis.

Knowledge mining extracts insights from large document collections.

Natural Language Processing (NLP) interprets human language.

Thus, based on the official Microsoft AI-900 study guide and Microsoft Learn, the correct and verified answer is An anomaly detection workload, since detecting unusual temperature fluctuations precisely fits this AI workload type.

According to the Microsoft Azure AI Fundamentals (AI-900) official learning path, the appropriate service for recognizing celebrities in images is Azure AI Vision (formerly Computer Vision). This service is part of Azure’s Cognitive Services suite and specializes in analyzing visual content using pretrained deep learning models. One of its built-in capabilities, as documented in Microsoft Learn: “Analyze images with Azure AI Vision”, includes object detection, face detection, and celebrity recognition.

The Azure AI Vision Analyze API can detect and identify thousands of objects, brands, and celebrities. When an image is submitted to the service, the model compares detected faces to a known database of public figures and returns metadata including celebrity names, confidence scores, and bounding box coordinates. This makes it ideal for applications that need to recognize well-known individuals automatically—such as media cataloging, content tagging, or entertainment apps.

The other options are incorrect:

A. Azure OpenAI Service provides generative AI and language models (like GPT-4), but it cannot analyze image content directly in the context of AI-900 fundamentals.

B. Azure Machine Learning is for custom model training and deployment, not a prebuilt vision recognition service.

C. Conversational Language Understanding (CLU) processes natural language input, not images.

Therefore, the correct service for identifying celebrities in images is D. Azure AI Vision.

In Microsoft’s Responsible AI framework, the Reliability and Safety principle ensures that AI systems perform consistently, safely, and as intended across diverse conditions — even when faced with incomplete, unusual, or unexpected data. Correctly handling unusual or missing values in a dataset directly demonstrates this principle, as it helps prevent faulty predictions, biased results, or unsafe system behaviors.

According to the Microsoft Learn Responsible AI module (from the AI-900 and AI-102 study paths), a reliable AI model should maintain its performance when encountering data anomalies. This includes validating inputs, managing missing or extreme values, and testing models to ensure they behave as expected in real-world scenarios. Such practices make AI systems robust and trustworthy, which aligns exactly with the Reliability and Safety principle.

The other Responsible AI principles address different concerns:

Inclusiveness: Ensures AI empowers and serves all users equitably.

Privacy and Security: Focuses on safeguarding personal data and preventing unauthorized access.

Transparency: Ensures that AI decisions are understandable and explainable to users.

While all principles are essential, managing data integrity and system stability—including how a model responds to missing or anomalous values—is primarily a matter of reliability and safety. It ensures the AI behaves predictably and minimizes risks of errors or unintended harm.

Therefore, the correct completion of the sentence is:

“Correctly handling unusual or missing values is an example of the application of the Reliability and Safety principle for Responsible AI.”