AI-102 Study

Vocab for AI-102

Person Group

A list or collection of people (persons)

Custom Scoring Script

Provides insight into model performance, but not direct or comprehensive. Would require additional development and maintenance, and might not capture all relevant metrics or provide real-time insights.

Location (Face and Emotion APIs)

Describes the rectangular coordinates of a face that is detected in an image. Includes the top, left, height, and width of a region in the image that displays a face.

Azure Data Factory

Can use workflows to orchestrate data integration and data transformation processes at scale. Build data integration, and easily transform and integrate big data processing and machine learning with the visual interface.

Analyze operation, “smart cropping”, Computer Vision API

Centers the thumbnail on the “region of interest”. If ‘smartCropping’ is false, crops to the center of the image.

Connect to an AKS node by using SSH

Add an SSH key to the node, and then you create an SSH connection

Values needed to make a call to Prediction API from client code

The Prediction URL and Prediction Key

Prediction API URL

Identifies the endpoint for the API connection

Prediction API Key

Authorized your app to access the service

Custom Vision Service Domain

Gives the classifier more information about the expected content of the images - Optimizes the classifier for a specific image type

Computer Vision API - Endpoint

Specifies the region you chose during sign up, the service URL, and a resource used on the request

Cognitive Service Contributor

This role provides permissions to manage cognitive services resources, including creating and managing deployments, keys, and other configurations.

Security Administrator

This role is more focused on managing security policies and configurations rather than managing cognitive services

Cognitive Services User

This role is typically for users who need to consume cognitive services but not manage them.

Lets you read and list keys of Cognitive Services

Security Manager

This role is more aligned with security management rather than managing cognitive services

Azure Stream Analytics Anomaly Detection

Offers built-in machine learning based anomaly detection capabilities that can be used to monitor the two most commonly occurring anomalies: temporary and persistent. Supports user-defined functions, via REST API, that call out to Azure Machine Learning endpoints

Azure Stream Analytics

A real-time analytics platform that can process large volumes of streaming data from IoT devices. Can filter, aggregate, and transform the data before storing it in a data warehouse.A

Azure Data Factory

Cloud-based ETL tool that can be used to orchestrate the data ingestion process. Can schedule data flows, transform data, and load it into a data warehouse

Azure Machine Learning Studio

Provides a drag-and-drop interface for building and training machine learning models.

Offers pre-built components for data preparation, feature engineering, and model evaluation.

Can be accessed from any web browser, making it accessible to developers on both Windows and Linux

Microsoft Visual Studio Code

A lightweight and versatile code editor that supports a wide range of programming languages.

Offers extensions for working with Azure Machine Learning, including tools for data preparation, model training, and deployment.

Can be used on both Windows and Linux, ensuring consistency across development environments.

Microsoft Visual Studio

Primarily focused on Windows development, so not best choice for using Azure Machine Learning

Azure Notebooks

Cloud-based Jupyter notebook environment that can be used for data analysis and machine learning, but might not offer the same level of integration with Azure Machine learning as Azure Machine Learning Studio or Visual Studio Code

Azure Data Lake Storage Gen2

• Scalability: Designed to handle massive datasets and can easily scale to accommodate an expected daily influx of >2 TB

• Cost-Effective: Offers a pay-as-you-go pricing model and provides optimized storage options for large datasets

• JSON Support: Supports storing data in various formats

• Performance: Designed for high-performance analytics and can handle large-scale data processing efficiently

Azure Video Indexer

Cloud application built on Azure Media Analytics, Azure Search, Cognitive Services.

Enables you to extract the insights from your videos using Video Indexer video and audio models:

• Visual Text Recognition (OCR): Extracts text that is visually displayed in the video

• Audio Transcription: Converts speech to text in 12 languages and allows extensions

• Sentiment Analysis: Identifies positive, negative, and neutral sentiments from speech and visual text

• Face Detection: Detects and groups faces appearing in the video

Computer Vision OCR

Detects text in an image using optical character recognition

Azure IoT Edge

Use a video stream to detect anomalies at the location

Azure Functions

Send the pictures and location information to Azure

Azure Logic Apps

Email a user the picture and location of the anomaly when an anomaly is detected

Cluster Autoscaler

Watch for pods in your cluster that can’t be scheduled because of resource constraints. When issues are detected, the number of nodes is increased to meet the application demand. Nodes are also regularly checked for a lack of running pods, with the number of nodes then decreased as needed.

Azure Stream Analytics

Uses an SQL-like query language that is very similar to Transact-SQL. Allows the data team to quickly adapt and use their existing skills to process and move data. Can write queries to filter, transform, and aggregate the streaming data before it’s written to storage.

Azure Notification Hubs

This service is for sending push notification, not for data processing and movement

Azure Event Grids

Serverless event routing service. Useful for reacting to events, but it’s not designed for continuous data processing and movement like Stream Analytics.

Ensure encrypted access to your published Azure Machine Learning web service

• Update DNS

  • You need to update your DNS records to point URL to the IP address or hostname of your deployed Azure Machine Learning web service

• Obtain an SSL certificate

  • An SSL certificate is essential for enabling HTTPS, which encrypts the communication between and the web service

• Update the Web Service

  • You must update the web service configuration, or deployment, to utilize the SSL certificate that you have obtained. This generally means configuring the web service to use HTTPS, and applying the certificate to the web service.

Term List API

The Text-Screen operation scans your text for profanity, and also compares text against custom and shared blacklists.

Face API Concepts

• Verification: Check the likelihood that 2 faces belong to the same person.

• Detection: Detect human faces in an image

• Identification: Search and identify faces

• Similarity: Find similar faces

• Grouping: Organize unidentified faces into groups, based on their visual similarity

Azure Functions to Process Data for OCR

Serverless computing units can be triggered by various events, including HTTP requests, timers, and messages from other Azure services

Enrolled

Enrollment status, indicates that an enrollment is ready for verification

None Intent

Random utterances that don’t map to any of your intents can be mapped to None.

Upload Data into QnA Maker Methods

1. Upload a text file

2. Provide the URL of a FAQ page

3. Manually enter data

Azure SQL Data Warehouse

Highly scalable data warehouse service that is optimized for large-scale data analytics and business intelligence workloads

• Scalability: Can handle large datasets like 200 TB and can be scaled up or down as needed to meet varying workload demands

• Performance: It is optimized for high-performance analytics queries, making it suitable for running complex machine learning experiments

• Cost-Effectiveness: Offers a pay-per-use pricing model, allowing only paying for resources your consume. Can optimize costs by scaling down the data warehouse during inactive periods

Troubleshoot node issue, need to connect AKS node by using SSH

1. Kubernetes Dashboard: Provides a visual interface to monitor node health, resource usage, and pod statuses

2. kubectl commands: Use commands like “kubectl describe node <node_name>”, “kubectl get pods -A”, and “kubectl logs <pod_name>” to gather information about node and pod health

3. Azure Portal: Check the node’s status, .logs, and metrics in the Azure portal

4. Azure Monitor: Use Azure Monitor to track node performance and identify potential issues

Azure Bot Service

Dynamically ask questions based on an uploaded image

Azure Computer Vision

Analyze and classify an image

Speech Translation API Translate output formats

audio/wav and audio/mp3

Authorization for QnA Service for a chatbot

In Application Settings for the bot, enter the:

• Knowledge base ID

• Host endpoint

• Endpoint key values

Classifier tags

Tell the classifier which characteristics apply to an image. Must add at least one tag for the image classification to work.

Model Versioning - Azure Machine Learning

Track different versions of your models, making it easier to compare performance, roll back to previous versions, or deploy specific versions to different environment

Model Logging - Azure Machine Learning

Enables you to log metrics, parameters, and other relevant information about your models, helping you understand their performance and make informed decisions.

Activities - Azure Machine Learning

Individual steps within a machine learning pipeline

Experiments - Azure Machine Learning

Track the training runs of your models

Deployments - Azure Machine Learning

Used to deploy models to endpoints

Pipelines - Azure Machine Learning

Orchestrate the entire machine learning workflow, including data preparation, training, and deployment

Web App Bot Template

Creates a Bot and an Azure Web App to host the Bot. Also configures the Bot to use a QnA Maker knowledge base. You will need to create the knowledge base yourself and connect it to the Bot.

Deployment Steps for AI Application Orchestrated by Kubernetes

1. Create a Kubernetes cluster: Provides the environment where your application will run

2. Create a container image file: Contains all the necessary components and dependencies for your application

3. Create an Azure Container Registry instance: Will store your container image and make it accessible to your Kubernetes cluster

4. Deploy your application to the Kubernetes cluster using the kubectl command-line tool

QnA Maker Knowledge Base

Consists of a set of question and answer (QnA) pairs and optional metadata that’s associated with each QnA pair

Value returned for category of text classification

Categories are rated with a value between 0 and 1, with values closer to 1 being more positive for the match.

ListId value in profanity check response

Identifier of the profanity word list that was used for the check

Azure Machine Learning Model Monitoring

Allows you to track model performance metrics like accuracy, precision, recall, and F1-score over time. Can also detect data drift and concept drift, which can impact model performance

Custom Monitoring

Can implement using tools like Prometheus and Grafana to collect and visualize relevant metricsA

A/B Testing

Deploy multiple versions of the model and compare their performance to identify the most accurate one

Regular Retraining

Regularly retrain your model on new data to maintain accuracy

Utterance, LUIS app

A phrase a user might use to interact with the application, such as, “Book a flight to New York.”

Machine Learning

The process of using available data to train a software model

Speech Translation API

Designed to do real-time speech translation for scenarios like in-person or remote translated communications and media subtitling.

Azure Data Factory

Retrieve data from file shares, Microsoft SQL Server databases, and Oracle databases that are in an on premises network

Azure Data Bricks

Process data stored in an Azure SQL Data Warehouse database

Azure Cognitive Services

Image-processing algorithms to smartly identify, caption, index, and moderate pictures and videosZ

Azure Data Factory

Service designed to allow developers to integrate disparate data sources. It is a platform somewhat like SSIS in the cloud to manage the data you have both on-prem and in the cloud. Provides access to on-premises data in SQL Server and cloud data in Azure Storage (Blob and Tables) and Azure SQL Database

Knowledge Mining

Azure Cognitive Search used knowledge mining to help retrieve information and extract insights through integration with other AI capabilities

Azure Machine Learning Layers

Refers to sophisticated pre-trained models, popular frameworks, productive services, powerful infrastructure, and flexible deployment.

Azure Blob vs Data Lake Storage

Data Lake will be a bit more expensive although they are in close range of each other.

Blob storage has more options for pricing depending upon things like how frequently you need to access your data (cold vs hot storage)

Row-Level Security Support

Supported by SQL Server, Azure SQL Database, and Azure SQL Data Warehouse

NV-Series Virtual Machine

Enables powerful remote visualization workloads and other graphics-intensive applications backed by the NVIDIA Tesla M60 GPU

The N-series is a family of Azure Virtual Machines with GPU capabilities. Ideal for compute and graphics-intensive workloads, helping customers to fuel innovation through scenarios like high-end visualization, deep learning, and predictive analytics

F-Series VMs

Feature a higher CPU-to-memory ratio. Example use cases include batch processing, web servers, analytics, and gaming.

A-Series VMs

Have CPU performance and memory configurations best suited for entry level workloads like development and testing.

BACON

System was designed to simulate the process of scientific discovery. It uses a set of heuristic rules to guide the search for patterns in data. It’s a classic example of an AI system that explores the role of heuristics in scientific discovery

Steps to Anonymize the data from IoT devices before data is setn to the IoT hub

1. Create a storage container

2. Create an Azure Stream Analytics Edge Job

3. Add the job to the IoT devices in IoT hub

Azure App Service Hybrid Connections

Azure websites and mobile services can access on-premises resources as if they were located on the same private network. Application admins thus have the flexibility to simply lift-and-shift specific most front-end tiers to Azure with minimal configuration changes, extending their enterprise apps for hybrid scenarios

Request-Response Service (RRS)

Designed for real-time synchronous interactions. When you want to consume a web service from Excel, you need immediate results returned to the spreadsheet. RRS provides this functionality

Batch-Execution Service

Used for asynchronous processing of large datasets. It is not suitable for real-time interactions with Excel.

API Key

Common method for authenticating and authorizing access to web services

Azure Managed Identity

While Azure managed identities are secure, they are primarily used for authentication between Azure resources.

Six Ethical Principles for AI

1. Fairness

2. Reliability & Safety

3. Inclusiveness

4. Privacy & security

5. Transparency

6. Accountability

Apache Hadoop

Particularly useful for big data and distributed processing. Can handle large amounts of data and can scale to accommodate the changing schemas of data files.S

Apache Spark

Designed for big data and can handle large-scale data processing. Known for its ability to handle dynamic schema changes and can be used for both batch and real-time data processing

A parallel processing framework that supports in-memory processing to boost the performance of big-data analytic applications

Speaker-dependent recognition

Involves training a speech recognition system on a specific speaker’s voice. This approach can significantly improve accuracy as the system learns to recognize the unique patterns and nuances of that individual’s speech

Foundational Elements of AI

Data, Cloud, and Algorithms

ND-Series VMs

Focused on training and inference scenarios for deep learning.

Microsoft Azure Table Storage

• Real-time querying: Azure Table Storage is designed to handle large volumes of data and allows for real-time querying as data streams into the solution. This is essential for IoT applications where data needs to be processed and analyzed immediately.

• Low Latency: Offers very low latency for loading data, making it suitable for scenarios where data needs to be ingested quickly.

• Scalability: Highly scalable and can handle large volumes of data without compromising performance.

• NoSQL Flexibility: Don’t need to define a schema upfront, making it flexible for handling unstructured or semi-structured data

Steps to deploy an Azure Machine Learning Studio image classification model as a containerized web service

1. Train the model

2. Create a container image

3. Register the container image

4. Get the http endpoint of the web service

Azure Application Insights Metrics for an Azure Bot

• The number of users interacting with the bot

• The number of messages interacting with the bot

• The number of messages on different channels received by the bot.

• The number of users and messages continuously interacting with the bot

Azure Monitor

A comprehensive solution for collecting, analyzing, and acting on telemetry from your cloud and on-premises environments

Bot Analytics

The analytics capabilities provided by the Azure Bot Service itself. Can provide some basic metrics, it is not as comprehensive or powerful as Azure Application Insights.

Azure Analysis Services

An enterprise-grade analytics as a service that lets you govern, deploy, test, and deliver your BI solution with ease. It is primarily used for data modeling and analysis, not for real-time monitoring or collecting KPI data from applications.

Steps to implement policy for changing access keys every 30 days

1. Generate new Key(s) in the Cognitive Service resources

2. Retrieve a token from the Cognitive Service endpoint

3. Update the custom application to use the new authorization

Strong Artificial Intelligence Approach

Aims to build machines that can truly reason and solve problems. These machines must be self-aware and their overall intellectual ability needs to be indistinguishable from that of a human being. Maintains that suitably programmed machines are capable of cognitive mental states.

Azure Cognitive Services Categories

• Vision

• Speech

• Language

• Knowledge

• Search

Monitor Accuracy of a Model

• Collects predictions: Capture predictions generated by the model for each input

• Compares predictions to ground truth: Compares these predictions with the actual ground truth values

• Calculates accuracy metrics: Computes relevant accuracy metrics such as accuracy, precision, recall, and F1-score

• Visualizes results: Presents the accuracy metrics in a clear and understandable format, such as charts or dashboards

Data Gateway for Azure Analysis Services

Can connect your Tabular Models hosted in Azure Analysis Services to your on-premises data sources through On-premises Data Gateway

Interactive Query

Provides in-memory caching and improved columnar storage engine for Hive queries