AI-900 Azure AI Fundamentals

When categorizing an image, the Azure AI Vision service supports two specialized domain models: celebrities and landmarks. Image types is an additional capability of the computer vision service, allowing it to detect the type of image, such as a clip art image or a line drawing. Both people_ and people_group are supported categories when performing image classification.

Azure Bot Service provides a user interface and connections to the different channels for Conversation AI agents or bots.

Options A is incorrect because the Azure Speech helps recognize and synthesize speech, recognize and identify speakers, translate live or recorded speech. It doesn’t provide a user interface for bots.
Option B is incorrect because the Bot Framework provides additional bots’ capabilities, but it relies on Azure Bot Service to provide a user interface for bots.
Option C is incorrect because the QnA Maker service provides knowledge base capabilities for bots, but it relies on Azure Bot Service to provide a user interface for bots.
Options E is incorrect because the Computer Vision service works with images, and It doesn’t provide a user interface for bots.

Step 1: Group customers based on behavior, age, and location

This requires Clustering (unsupervised learning) because we're grouping similar customers without predefined labels
The model will discover natural groupings in the customer base

Step 2: Predict whether each group will spend >$1000

This is a Binary Classification problem (supervised learning) because:
It's a yes/no prediction (will or won't spend >$1000)
It has a specific label we're trying to predict

Why other options are incorrect:

A) Supervised Classification first wouldn't make sense because we need to group customers before making predictions
C) Multiclass Classification alone wouldn't address the grouping requirement
D) Regression would predict a specific numeric value, which isn't needed here; we only need a yes/no prediction

Key points:

Multiclass classification predicts "mutually exclusive" labels
Each input can only be assigned to one class
In real-world scenarios, items might legitimately belong to multiple categories

System messages should be used to set the context for the model by describing expectations. Based on system messages, the model knows how to respond to prompts. The other techniques are also used in generative AI models, but for other use cases.

Key points:

Transformers learn semantic relationships between words
They determine probable sequences based on training data
They use encoder blocks for vocabulary representation
They use decoder blocks for generating sequences

Why option C is correct:

The model only understands statistical relationships between words
It can generate "probable" sequences that make linguistic sense
It has no actual understanding of medical practice or consequences
It's working exactly as designed, but this reveals its fundamental limitation

Why other options are incorrect:

A) More training data wouldn't give the model real medical understanding
B) The encoder is likely working properly if responses are technically correct
D) The architecture isn't the issue; it's the inherent limitation of text prediction

The key insight is that transformer models, despite their sophistication, only process semantic relationships and probabilities - they don't truly understand the real-world implications of the text they generate.

The Split Data module is particularly useful when you need to separate data into training and testing sets. Use the Split Rows option if you want to divide the data into two parts. You can specify the percentage of data to put in each split, but by default, the data is divided 50-50. You can also randomize the selection of rows in each group, and use stratified sampling.

Key points:

Regression models predict numeric labels (house prices in this case)
The features include both numeric (square footage, bedrooms, age) and non-numeric (location) data
Feature engineering would be needed to convert location into a numeric format (like coordinates or encoded categories)

Why other options are incorrect:

A) Not all features are numeric (location isn't), so it won't work perfectly without modification
B) Having non-numeric features doesn't disqualify using regression; they just need to be properly encoded
D) Limiting to only numeric features would ignore valuable predictive information from other features

The key insight is that regression can handle this prediction task, but proper data preparation (feature engineering) is needed to convert non-numeric features into a format the model can use.

Tokenization is part of speech synthesis that involves breaking text into individual words such that each word can be assigned phonetic sounds. Transcribing is part of speech recognition, which involves converting speech into a text representation. Key phrase extraction is part of language processing, not speech synthesis. Lemmatization, also known as stemming, is part of language processing, not speech synthesis.

Key points:

Binary classification has two possible outcomes (positive/negative)
The model has high accuracy for negative cases (95% correct)
The model has lower accuracy for positive cases (60% correct)
In medical contexts, missing disease cases (false negatives) can be life-threatening

Why other options are incorrect:

A) Overall accuracy isn't the most important metric in medical diagnosis
B) The model actually performs well with healthy patients (95% correct)
D) Processing time isn't mentioned and isn't relevant to the accuracy issue

The key insight is that in medical applications, missing actual disease cases (false negatives) is typically more dangerous than incorrectly flagging healthy patients (false positives), as it could lead to untreated conditions with serious health consequences.

A clustering algorithm is an example of unsupervised learning, which groups data points that have similar characteristics without relying on training and validating label predictions. Supervised learning is a category of learning algorithms that includes regression and classification, but not clustering. Classification and regression algorithms are examples of supervised machine learning.

Identifying potential harms is the first stage when planning a responsible generative AI solution.

During pre-processing, you need to work with data to select features that influence the label prediction. In this problem, features are the engine characteristics (power or volume), seat comforts, etc. They could help the ML model to predict the success of the new car model. Maybe the sunroof is not essential for predicting the label, and we need to discard this feature from the final set of features that we will use for model training.

In short, Feature selection helps us to narrow down the features that are important for our label prediction and discard all features that don’t play or play a minimal role in a label prediction. As a result, our trained model and prediction will be more efficient.

All other options are incorrect because they are parts of the different data processing events that are irrelevant to the pre-processing (Training set selection or Data for model evaluation selection) or too generic (Data selection or Data Classification).

Named Entity Recognition (NER) is a Text Analytics service that helps identify entities in the text and group them into different entity categories, like person, organization, location, event, etc.

You should expect three recognized named entities with the category “Location” in the API response- “headquarters,” “Paris,” and “Eiffel tower.”

The accountability principle states that AI systems are designed to meet any ethical and legal standards that are applicable. The system must be designed to ensure that privacy of the healthcare data is of the highest importance, including anonymizing data where applicable. The fairness principle is applied to AI systems to ensure that users of the systems are treated fairly. The inclusiveness principle states that AI systems must empower people in a positive and engaging way.

Accountability defines the framework of governance and organizational principles, which are meant to ensure that AI solutions meet ethical and legal standards that are clearly defined. The other answer choices do not define the framework of governance and organization principles, but provide guidance regarding the ethical and legal aspects of the corresponding standards.

Generative AI models offer the capability of generating images based on a prompt by using DALL-E models, such as generating images from natural language. The other AI capabilities are used in different contexts to achieve other goals.

Transparency provides clarity regarding the purpose of AI solutions, the way they work, as well as their limitations. The privacy and security, reliability and safety, and accountability principles focus on the capabilities of AI, rather than raising awareness about its limitations.

Live speech translation involves the following sequence of the services during the real-time audio stream- Speech-to-Text -> Speech Correction -> Machine Translation -> Text-to-Speech.
Options A is incorrect because the Speech Recognition is not involved in this process. You can define from and to translation languages in Speech Translation service settings.
Option E is incorrect because we are not using Text Analysis services in this case.

Removing stop words is the first step in the statistical analysis of terms used in a text in the context of NLP. Counting the occurrences of each word takes place after stop words are removed. Creating a vectorized model is not part of statistical analysis. It is used to capture the semantic relationship between words. Encoding words as numeric features is not part of statistical analysis. It is frequently used in sentiment analysis.

With only 5% of data held back for validation, and two possible outcomes (pass/fail):

There might not be enough examples of each class in the validation set to meaningfully evaluate the model
This is especially important when classes are imbalanced
Small validation sets can lead to unreliable performance estimates

Linear regression is a machine learning algorithm module used for training regression models. The Clean Missing Data module is part of preparing the data and data transformation process. Select Columns in Dataset is a data transformation component that is used to choose a subset of columns of interest from a dataset. Evaluate model is a component used to measure the accuracy of trained models.

Modern image classification solutions are based on deep learning techniques. Semantic segmentation provides the ability to classify individual pixels in an image depending on the object that they represent. Both linear regression and multiple linear regression use training and validating predictions to predict numeric values, so they are not part of image classification solutions.

Image classification is part of computer vision and can be used to evaluate images from an X-ray machine to quickly classify specific bone fracture types. This helps improve diagnosis and treatment plans. An image classification model is trained to facilitate the categorizing of the bone fractures. Object detection is used to return identified objects in an image, such as a cat, person, or chair. Conversational AI is used to create intelligent bots that can interact with people by using natural language. Facial detection is used to detect the location of human faces in an image.

Entity Linking identifies and disambiguates the identity of entities found in a text. Key phrase extraction is not used to extract entities and is used instead to extract key phrases to identify the main concepts in a text. Named entity recognition cannot provide a link for each entity to view further information. Text translation is part of the Azure AI Translator service.

Azure AI Custom Vision is an image recognition service that allows you to build and deploy your own image models. The Azure AI vision service, Azure AI Face service, and Azure AI Language service do not provide the capability to train your own image model

Classification is used to predict categories of data. It can predict which category or class an item of data belongs to. In this example, a machine learning model trained by using classification with labeled data can be used to determine the type of bone fracture in a new scan that is not labeled already. Featurization is not a machine learning type. Regression is used to predict numeric values. Clustering analyzes unlabeled data to find similarities in the data.

Key points:

Attention quantifies the strength of relationships between tokens
It examines the sequence of text tokens
It measures how strongly words are related to each other

Why option B is most challenging:

"it" could refer to either "cat" or "mat"
Determining the correct reference requires understanding context
The model must analyze relationships across multiple words and concepts
Resolving pronouns (like "it") requires understanding more than just direct word relationships

Why other options are less challenging:

A) "cat" and "sat" have a direct subject-verb relationship
C) "the" and "mat" have a clear article-noun relationship
D) "sat" and "on" have a straightforward verb-preposition relationship

The key insight is that while attention can measure relationships between tokens, ambiguous references require more complex analysis of context and meaning, making pronoun resolution particularly challenging.

etecting objects identifies common objects and, for each, returns bounding box coordinates. Image categorization assigns a category to an image, but it does not return bounding box coordinates. Tagging involves associating an image with metadata that summarizes the attributes of the image, but it does not return bounding box coordinates. OCR detects printed and handwritten text in images, but it does not return bounding box coordinates

Key phrase extraction is used to extract key phrases to identify the main concepts in a text. It enables a company to identify the main talking points from the support question data and allows them to identify common issues. Named entity recognition can identify and categorize entities in unstructured text, such as people, places, organizations, and quantities. The Azure AI Speech service, Conversational Language Understanding, and Azure AI Bot Service are not designed for identifying key phrases or entities.

In a regression machine learning algorithm, features are used to generate predictions for the label, which is compared to the actual label value. There is no direct comparison of features or labels between the validation and training datasets.

Key points:

The process is sequential
Each stage builds on the previous one
Missing something in Stage 1 affects all subsequent stages

Why option A is correct:

Stage 1 (Identify potential harms) is foundational
If reputational harm wasn't identified as a potential issue:
It wouldn't be measured in Stage 2
No mitigation strategies would be planned in Stage 3
No operational guidelines would address it in Stage 4

Why other options are less critical oversights:

B) You can't properly measure what you haven't identified as a potential harm
C) You can't mitigate what you haven't identified and measured
D) Operational guidelines can't address unidentified issues

The key insight is that failing at Stage 1 creates a cascade of problems throughout the entire process. If you don't identify a potential harm at the beginning, you can't properly address it in any subsequent stage.

The deep learning models are characterized by their deep architectures, which consist of multiple layers of interconnected neurons. These layers allow the model to learn hierarchical representations of data, where each layer extracts increasingly abstract features from the input data. Training these models often requires processing large amounts of data through numerous layers, leading to longer training times compared to simpler machine learning algorithms.

Option A is INCORRECT because the deep learning algorithms typically require large amounts of training data. The use of data for predictions is not a contributing factor to the longer training times.

Option B is INCORRECT because while this is a characteristic of deep learning, it’s not directly related to longer training times. Deep learning algorithms can create high-level features autonomously, but this doesn’t necessarily lead to longer training times.

Option C is INCORRECT because this option describes a characteristic of machine learning, not deep learning. Deep learning algorithms often tackle problems on an end-to-end basis rather than breaking them into smaller steps.

Option D is INCORRECT because in deep learning, features are typically learned from the data automatically, without requiring user input. This involvement of users in feature engineering is more common in traditional machine learning approaches.

The AI system must be designed to ensure that biased decision making is avoided and not based on factors such as ethnicity and gender. The system will consider salary, payment history, and credit utilization. These are standard metrics.

In a regression machine learning algorithm, a validation set contains known feature and label values.

The price of the house is the label you are attempting to predict through the machine learning model. This is typically done by using a regression model. Floor space size, number of bedrooms, and age of the house are all input variables for the model to help predict the house price label.

Entity Linking, PII detection, and sentiment analysis are all elements of the Azure AI Service for Azure AI Language. Azure AI Vision deals with image processing. Azure AI Content Moderator is an Azure AI Services service that is used to check text, image, and video content for material that is potentially offensive.