Free Microsoft Azure AI Fundamentals (AI-900) Questions and Answers — Questions and Answers

Question 1: In this computer vision challenge, each pixel in an image is categorized according to the object to which it belongs.

Optical Character Recognition
Image Analysis
Object Detection
Semantic Segmentation (Correct answer)

Correct answer: Semantic Segmentation

Semantic segmentation is a computer vision task where each pixel in an image is classified with a label corresponding to a predefined object class. Unlike object detection, which draws bounding boxes, semantic segmentation provides a pixel-level understanding of the image, delineating the exact boundaries of each object.

Question 2: With the aid of this computer vision approach, text may be found and read within images.

Image Analysis
Optical Character Recognition (Correct answer)
Object Detection
Semantic Segmentation

Correct answer: Optical Character Recognition

Optical Character Recognition (OCR) is a computer vision technique specifically designed to detect and extract text from images, whether they are scanned documents, photos, or screenshots. It converts various types of documents into editable and searchable data, making text within images readable by machines.

Question 3: Which of these can be utilized to develop conversational AI?

Azure Speech Maker
Q&A Maker (Correct answer)
Azure Chat Functions
Semantic Segmentation

Correct answer: Q&A Maker

Azure QnA Maker is a cloud-based Natural Language Processing (NLP) service that allows users to create a conversational layer over their data. It enables the creation of knowledge bases from existing content and provides an API to integrate question-and-answer capabilities into applications and bots, effectively building conversational AI.

Question 4: Which of the following resources cannot be produced using Azure Machine Learning Studio?

Compute Balancers (Correct answer)
Compute Instances
Inference Clusters
Compute Clusters

Correct answer: Compute Balancers

Azure Machine Learning Studio provides various compute resources for training and deploying models, including Compute Instances (managed workstations), Compute Clusters (scalable training environments), and Inference Clusters (for deploying models). 'Compute Balancers' is not a standard resource type offered within Azure Machine Learning Studio.

Question 5: Which would be utilized to determine the category to which an item belongs?

Neural Network
Clustering
K-means
Classification (Correct answer)

Correct answer: Classification

Classification is a supervised machine learning task used to predict a categorical label for a given input. The goal is to assign an item to one of several predefined categories or classes based on its features, such as determining the type of an object or categorizing an email.

Question 6: What method is used to group comparable entities based on features?

Clustering (Correct answer)
Classification
Neural Network
K-means

Correct answer: Clustering

Clustering is an unsupervised machine learning technique used to group similar data points together based on their inherent characteristics or features. Unlike classification, clustering does not rely on predefined labels but rather discovers patterns and structures within the data to form natural groupings.

Question 7: What resource kinds are offered as part of the computer vision service?

Computer Key and Computer Endpoint
Computer Insights and Computer Services
Computer Vision and Computer Cluster
Computer Vision and Computer Services (Correct answer)

Correct answer: Computer Vision and Computer Services

When creating an Azure Computer Vision resource, you provision a 'Computer Vision' resource type, which provides the API endpoint and keys to access the service. While 'Computer Services' is not a formal resource type, it broadly refers to the capabilities and functionalities offered by the Computer Vision service within the Cognitive Services umbrella.

Question 8: What are the two specialized domain models of computer vision?

Marker and People
Stars and Places
Celebrities and Landmarks (Correct answer)
Background and Landmarks (Correct answer)

Correct answer: Celebrities and Landmarks

Azure Computer Vision offers specialized domain models that are pre-trained for high accuracy on particular types of content. Two prominent specialized domain models are 'Celebrities,' used for identifying famous people, and 'Landmarks,' used for recognizing well-known natural or man-made structures.

Question 9: Which of these metrics is not applied to Custom Vision model training?

Recall
F1 Score (Correct answer)
Precision
Mean Average Precision

Correct answer: F1 Score

While F1 Score is a common metric in machine learning, Azure Custom Vision's portal and documentation primarily emphasize Precision, Recall, and Mean Average Precision (mAP) as the key performance indicators displayed for model training results. Therefore, F1 Score is not typically presented as a primary, directly 'applied' metric within the service's user interface compared to the others.

Question 10: Which image format in the Face Service is not supported?

BMP
PNG
JPEG
AI (Correct answer)

Correct answer: AI

The Microsoft Azure Face Service supports common raster image formats like JPEG, PNG, and BMP for facial analysis. The 'AI' format typically refers to Adobe Illustrator vector graphics, which are not pixel-based and therefore not directly supported by the service for image processing tasks like face detection or recognition. The service requires standard photographic image formats to extract facial features.

Question 11: All of the information listed here, with the exception of:

Areas
Words
Lines
Areas (Correct answer)
Regions

Correct answer: Areas

This question, likely referring to document processing services like Azure Form Recognizer, asks what information is *not* typically listed or extracted. Services like Form Recognizer are designed to extract structured information such as individual words, lines of text, and logical regions (like paragraphs or fields). 'Areas' is a very general term and doesn't represent a specific, actionable type of information extracted in the same granular way as words, lines, or defined regions.

Question 12: Which translation model is used by translation text?

SVM
Cognitive Text
Neural Machine Translation (Correct answer)
Neural Network

Correct answer: Neural Machine Translation

Modern text translation services, including Azure's Cognitive Services Text Translation, primarily utilize Neural Machine Translation (NMT). NMT models leverage deep learning to process entire sentences and their context, leading to significantly more fluent, accurate, and natural-sounding translations compared to older statistical or rule-based methods. This advanced approach allows for a deeper understanding of language nuances.

Question 13: Which of the following does not fall within the category of LUIS application intents?

RegEx
List
Machine-Learned
Filter (Correct answer)

Correct answer: Filter

In LUIS (Language Understanding Intelligent Service), intents categorize user utterances based on their purpose. Intents can be defined and enhanced using various methods, including Regular Expressions (RegEx) for pattern matching, List entities for predefined sets of values, and Machine-Learned entities for context-aware extraction. 'Filter' is an action or operation that might be performed *after* an intent is identified, rather than a type of intent or a method for defining one within LUIS.

Question 14: Which classification doesn't fall within machine learning algorithms?

Clustering
Normalization (Correct answer)
Regression
Classification

Correct answer: Normalization

Clustering, Regression, and Classification are fundamental categories of machine learning algorithms, each addressing different types of problems (unsupervised, supervised continuous, and supervised discrete, respectively). Normalization, however, is a data preprocessing technique used to scale numerical features to a standard range, which helps algorithms perform better. It is a data transformation step, not a machine learning algorithm itself.

Question 15: Which one is a multiclass classification algorithm?

K-means
Decision Forest (Correct answer)
Anomaly Detection
K-nearest Neighbor

Correct answer: Decision Forest

A multiclass classification algorithm is used when the target variable can belong to more than two distinct categories. Decision Forest is an ensemble learning method that constructs multiple decision trees and combines their predictions, making it highly effective for complex multiclass classification tasks. K-means is a clustering algorithm, and Anomaly Detection identifies outliers, while K-nearest Neighbor can be used for classification but Decision Forest is a prominent example of a multiclass classifier.

Question 16: This is done to draw attention to the strongest trend in a dataset:

K-means
Reinforcement Learning
Classification Algorithm
Principal Component Analysis (Correct answer)

Correct answer: Principal Component Analysis

Principal Component Analysis (PCA) is a dimensionality reduction technique that identifies the principal components, which are new variables that capture the maximum variance in the dataset. By transforming the data into these components, PCA effectively highlights the strongest trends and underlying structure, making it easier to visualize and analyze complex datasets. It helps in simplifying data while retaining most of its important information.

Free Microsoft Azure AI Fundamentals (AI-900) Questions and Answers

In this computer vision challenge, each pixel in an image is categorized according to the object to which it belongs.