AI 900

Data Ingestion

is the process in which unstructured data is extracted from one or multiple sources and then prepared for training ML models.

Value Clipping

is the process of detecting outliers to remove or replace them with a mean, constant, or substitute values.

Data Export

is the process of saving machine learning results, intermediate data or working data into the local or cloud storage destinations.

Azure ML STudio

A collaborative visual workspace where you can build, test, and deploy machine learning solutions by using prebuilt machine learning algorithms and data-handling modules. Supports code AND no code experiences. It does support Jupyter notebooks that use Python SDK but it does NOT support the use of C# or .NET (use visual studio for those using ML.NET SDK)

Azure Kubernetes Service (AKS)

A complete orchestration service for containers including autoscaling, coordinated app upgrades, and service discovery across multiple containers. PaaS storage offering. Supports compute targets for container model deployments.

Personal Digital Assistant (PDA)

A conversational AI solution that provides management, retrieval and update of users' personal or business information to keep them informed and productive. It can run across devices and platforms with access to electronic calendars, e-mail, contact lists and other applications to enable personalized help with routine tasks. EX: use Siri or Cortana to text a friend.

Feature Engineering

is a process of creating new features from within provided input datasets, which can help to better represent relationships and correlations between the input data and expected outcome. This solution also improves the quality and performance of the trained model and increases its predictive power.

Coefficient of determination

A metric used to evaluate a regression machine learning model. Also known as R2 or R-squared, is a measure of the variance from the mean in its predictions. The closer this metric is to 1, the better the model is performing.

Relative absolute error (RAE)

A metric used to evaluate a regression machine learning model. This metric is a measure of the difference between predictions and actual values. Calculating total absolute error. The closer this metric is to 0, the better the model is performing.

Normalize Data Module

A module that can be added to the training pipeline in Azure Machine learning designer that adjusts the values in the numeric columns so that all numeric columns are on a similar scale between 0 and 1. A data set that has features in different scales can bias the model towards that feature, to mitigate the bias in the model you transform the numeric features to use the same scale through this module.

Select Columns Dataset Module

A module that can be added to the training pipeline in Azure Machine learning designer that removes columns from the data set and create a smaller data set. This module does NOT change the data values.

Clean Missing Data Module

A module that can be added to the training pipeline in Azure Machine learning designer that removes, replaces, or in furs missing values in the data set. This module can also remove empty rows from the data set. Missing data can limit the accuracy and effectiveness of predictions, but this module does NOT adjust the scale of the data

Azure ML Designer

A workspace only available with an Enterprise edition. You can connect modules directly together but you CANNOT connect datasets. Like data sources, datasets have only output ports and can only connect to modules. Pipelines published to a pipeline endpoint can be used to train models, process new data but CANNOT be sent from a pipeline endpoint in real time.

Channel

An action when creating the Azure bot framework. You can add this to make the bot available on the related platforms like facebook, email, teams etc. Default is Web

Image Classification

A ML model and custom vision service that is the process of applying class or category labels to images according to their visual characteristics. You can use a pre-trained version of this AI service to train your own classification with a set of custom images. EX: analyzing car damage, identifying products on a warehouse shelf or perform medical diagnosis on MRI Scan

Face API Service

A service that can detect the angle a head is posed at. Concerned with the details of faces.

Video Indexer Service

A service that extracts insight from video images.

Skill

Another bot that can an action when creating the Azure bot framework. This is where a bot can perform tasks for another bot. You can add this when you want to reuse or extend a bot. An Azure bot can both consume other bots or be consumed itself. JSON file. A root bot also consumes this and can be user facing.

Speech-to-Text

An API that can be used synchronously (real time) or asynchronously (batch). Batch includes large volumes of audio recordings stored in Azure Blob.

Image Classification Model

Assessing the damage to a vehicle from a photograph is an example of?

Tab Separated Value (TSV)

File format to upload a chit-chat personality to a QnA Maker knowledge base.

Language Understanding Intelligence Service (LUIS)

Can analyze user utterances (text in the users own words) to determine their intention. Has built-in support for multiple languages, including, but very limited. However it cannot be used as a proper text to text translation service in near real time. Can pair with QnA maker to process texts received by chatbots and determine user intentions. 2 main resources in this type of app: ID and prediction key

Voice Assistant

Can perform speech to text and text to speech processing for bots created with the Azure Bot service

Entity Recognition

Finds the use of people, places, or organizations, and other known items contained in a piece of text. The entities it returns are from a set of known entities

Entity Recognition

Detects the use of people, places, organizations, and other known items from a piece of TEXT

Algorithm

Discovers patterns in the data when a model is trained. Is a set of commands or instructions that a computer typically executes in a sequence. In machine learning specifically, these operate on the provided dataset to produce relevant classification, regression, or clustering. They are NOT part of the input dataset.

Key phrase extraction

Evaluates a piece of text and identifies the core talking points or popular mentions contained in text. You can extract these from website reviews and then add these tags to allow website users to filter reviews. Works better on larger amounts of text.

Conversational AI

Examples include: webchat bot, PDA, and telephone voice menus.

Responsible AI Principles

Fairness, Inclusiveness, Accountability, Transparency, Reliability and Safety, Privacy and Security

Regression Model

In a machine learning scenario, this model is used to make predictions in a continuous form by learning from labeled historical data to predict or forecast new values. Can use the Coefficient of determination metric as well as R2 and RMSE Ex: Predict house price based on the location, # of rooms, and other factors.

Classification Model

In a machine learning scenario, this model is used to make predictions in a non-continuous form by learning from labeled data to create new observations. Can use the Precision metric, recall, and AUC metric. Ex: Checking if new email messaged are spam or not.

Clustering Model

In a machine learning scenario, this model iteratively groups cases in an unlabeled dataset into a _______ that has similar characteristics. Can use the Average distance to center metric and max distance to center. Ex: Grouping together online shoppers with similar traits for targeted marketing.

Feature Column

In machine learning, these columns are a descriptive attribute used to train classification models to predict a class or category of the outcome based off of this attribute.

Label Column

In machine learning, these columns are the outcomes that the model needs to predict or forecast.

Uploading a PDF and Manually adding pairs

In which 2 ways could you natively populate a QnA Maker knowledge base?

Driving autonomous vehicles. Think of Tesla recognizing other cars and people and grouping those pixels to an object. The AI system must have necessary perception of the surrounding environment

In which scenario would you use semantic segmentation?

Clustering

Is a machine learning method of grouping data points that possess similar traits

Knowledge Mining

Is a powerful AI tool to help you deeply understand and explore vast amounts of information to get better insight and uncover hidden relationships and patterns in your data. NOT designed to detect abnormalities.

Auto ML model

Is a process in which the best machine learning algorithm to use for your specific data is selected for you. This service tries different algorithms and tuning parameters in parallel while training a model and stops once it reaches predefined exit criteria. You can register a File and Tabular dataset to assist in training this model. you can also create a dataset from the OUTPUT of a sql query, a dataset from files inside a folder, or a dataset from files in a container but not the container/Folder/or query itself.

Prediction workload

Is the process of using machine learning to analyze the patterns in provided data and associates historical patterns to infer possible outcomes. NOT designed to detect abnormalities.

Anomaly Detection

Is the process of using machine learning to monitor and detect unexpected items or events in time series data that differ from the norm. It enables preemptive action to be taken before a problem occurs and analyzes data over time. Does NOT predict when problems will occur.

Speech Synthesis Markup Language (SSML)

Lets you improve the quality of speech synthesis by fine tuning pitch, pronunciation, speaking rates, and other parameters of the text-to-speech output. Format is XML as per the World Wide Web consortium standard.

Azure Container Instances (ACI)

Offers the fastest and simplest way to run a container in Azure. You don't have to manage any virtual machines or configure any additional services. It is a PaaS storage offering that allows you to upload your containers and execute them directly. Supports compute targets for container model deployments.

facial detection

The Detect API detects human faces in an image and returns the rectangle coordinates of their locations. It also returns a unique ID that represents the stored face data, which is used in later operations to identify or verify faces. Wearing sunglasses can negatively affect performance, and slower shutter speed does NOT improve performance

Area under the curve (AUC)

This metric is used when evaluating classification models. It measures the area underneath a curve that represents true positive rate over true negative rate. This metric ranges between 0-1. Values closer to 1 indicate that the models performing better.

Precision metric

This metric is used when evaluating classification models. This metric is the number of true positives divided by the sum of true positives and false positives. This metric is scored between 0-1. Values closer to 1 are better.

Average Distance to Cluster Center

This metric is used when evaluating clustering models. This metric is a measure of how focused the clusters are. The lower, the better.

create an inference pipeline from the training pipeline

What should you do after training your model and prior to deploying as a web service?

Score the model - applies new data to the trained model to generate predictions

What should you do to measure the accuracy of a trained ML model?

Evaluate the Model - accuracy of performance

What should you do to measure the accuracy of the predictions and assess model fit?

Kubernetes and container instances

Which 2 compute targets can you deploy ML model as a web service?

QnA Maker and Language Understanding Intelligence Service (LUIS)

Which two Azure Cognitive services should you combine so that your chatbot can determine user's intentions and find answers from a custom knowledge base?

Bot Framework CLI tools

a bot service template used to manage bots and related services and are used in DevOps pipelines when deploying bots in enterprises

Basic edition Azure ML

a code first experience, a workspace where you can create, run, and publish models but only using SDK. Does NOT allow the use of ML designer.

Azure Open Dataset

a curated dataset made available on Azure that you can import into you ML learning model.

Validation dataset

a dataset used to provide an evaluation of the model to see if the model can correctly predict, or classify, using data not seen before. Used to tune the ML data model. Is a sample of data held back from the training of the ML model and it helps to get an unbiased evaluation of the model when tuning. Use after training but before final testing.

Testing dataset

a dataset used to provide final unbiased evaluation of the model. An independent sample of data and is used once a model has been completely trained with the training and validation datasets.

Training dataset

a dataset used to train an ML model for future prediction. It is the largest sample of data used when creating an ML model.

Domain

a requirement to a LUIS prebuilt model type that contains intents, utterances, and entities. Common utterance can be a control for a smart devices with intents such as TurnOn and TurnOff and an entity such as "light"

Entity

a requirement to a LUIS prebuilt model type trained to extract specific data. Ex: extract temperature information

Intent

a requirement to a LUIS prebuilt model type, contains utterances but NOT entities. You can also add these from the prebuilt domains without adding the entire domain model. The todo.Confirm _____ contains utterances that confirm that a task should be performed.

Reliability and safety

a responsible AI principle requires AI systems to perform safely without any substantial risk to human life

Transparency

a responsible AI principle that dictates AI based solutions to be understandable. Users should be aware the purpose of the AI based system , how it operates, it's scope, and limitations.

Inclusiveness

a responsible AI principle that focuses on empowering everyone. AI should provide benefits to all parts of society regardless of gender, physical ability, or any other factor.

Privacy and Security

a responsible AI principle that requires the AI based system to respect privacy and be secure. Systems operate on high volumes of data, including PII that should not be disclosed and this data is subject to privacy rules.

Bot Framework Composer

built on the Bot Framework SDK, is an open-source IDE for developers to author, test, provision, and manage conversational experiences. It provides a powerful visual authoring canvas enabling dialogs, language-understanding models, QnAMaker knowledge bases, and language generation responses to be authored from within one canvas and crucially, enables these experiences to be extended with code for more complex tasks such as system integration. It can publish bots to Azure Web and Azure functions. Available as a desktop application for Windows, macOS, and Linux. If the desktop app isn't suited to your needs

Azure IoT

enables management of devices and bi directional communication between them and apps. CANNOT deploy ML here. (device management not analysis)

Confusion Matrix

generated for a classification model and shows the number of true and false results for each class.

Enterprise edition Azure ML Designer

has all the features of the basic edition plus the ability to create and publish models with no code

Bot Service templates

includes five templates to help you get started with building bots. These templates provide a fully functional bot out of the box to help you get started quickly. When you create a bot, you choose a template and the SDK language for your bot. Types: basic, form, QnA, Proactive, language understanding

Azure Bot Emulator

is a desktop application that allows developers to test and debug on their local computer

Text Analytics Service

A cognitive service including: sentiment analysis, language detection, entity deduction, and key phrases. Can be used with PBI to analyze text. Has a rest API that enables .NET C# Java Python Ruby and Go

Telephone voice menu

A conversational AI solution that can reduce workload on the human operators, provide generic instructions to the customers, automatically transfers calls to the relevant teams or manage the waiting queue, supporting business operations even during non working hours. Integrated with a phone system and supports interaction through the phone keypad or speech recognition.

Webchat bot

A conversational AI solution that can utilize cognitive services and knowledge bases to conduct real-time conversations with humans via web channels and is typically integrated with web-enables apps.

Azure Service Bus

A message broker/queueing system that enables reliable and secure asynchronous transfer of data and state. (pre analysis still ingesting)

Optical Character Recognition (OCR) API

A solution that can translate a scanned document image into text that can be edited and formatted. Extracts small amounts of printed or handwritten text from an image. Can recognize individual shapes as letters, numerals, punctuation, and other elements of text. Can extract details from invoices that have been sent or scanned. Then can validate against the expected details in your financials. Ex: can identify and read a license plate, but NOT assess damage to a vehicle.

Language Modeling

A type of LUIS, Aims to interpret the intent from a text command statement and extract key information to discover the overall meaning from text NOT emotion. CANNOT detect the language as it only does one at a time. EX: converting command into smart actions

Natural Language Processing (NLP)

An AI solution that supports applications that can see, hear, speak with, and understand users. Using text analytics, translation, and language understanding services, Microsoft Azure makes it easy to build applications that support natural language. Can be provisioned via a language resource or a cognitive service resource. EX: speaker recognition, speech translation, sentiment analysis, speech synthesis

Text-to-Speech

An API that converts text into synthesized speech. You can choose from standard neural or customized voices.

Speech Translation

An Azure cognitive service and NLP workload that can provide multi lingual English translation in your time. However, this service expects input to be in audio format, i.e. as a file or audio stream. It does not support text as an input, and that is why it CANNOT be used to translate messages in a custom chat application.

Speaker Recognition API

An Azure cognitive service and NLP workload that gives your app the ability to recognize and identify individual speakers.

Speech Synthesis

An Azure cognitive service and NLP workload with the ability to generate spoken output by converting text into audio speech. Can support people with vision impairment by converting input text to human-like speech.

Sentiment Analysis

An Azure cognitive service and NLP workload, this tool returns sentiment labels and scores for the entire document, and evaluates text and then return sentiment labels such as negative, neutral or positive. Works better on smaller amounts of text. Determines the emotion in the statement. It scores at both the sentence and document levels. As part of this the levels are applied to text and at the sentence and document labels giving a confidence score for each label: positive, negative or neutral. Confidence scores range from 0-1.

Text Translator

An Azure cognitive service that can enable text translation in your chat application in near real time. Supports translation into and out of more than 70 languages. Can translate from one language into multiple simultaneously and recognizes variants like UK english

Personalizer

An Azure cognitive service that can learn about user preferences from their collective real-time behavior, and can choose then the best items, such as specific products to recommend to each user.

Form Recognizer

An Azure cognitive service that in just text from forms and extracts key value pairs in table data from them. An API that extracts data from a doc and provides a GUI to visualize the data in a table like format. It is useful for the extraction of key information from invoices, sales receipts, business cards etc. Cannot enhance the functionality of your custom chat application. Prebuilt model CANNOT extract info from sales receipts in French

Custom Vision Service

An Azure cognitive service to that allows you to build and deploy image classifier trained on your custom set of images and labels. You can create your own labels to tag images with. You cannot add your own labels for any other service. Two different project types: Classification and Object Detection

Computer Vision Service

An Azure cognitive service with a rich set of image processing functionalities to detect objects, brands, or faces, describe image content or generate thumbnails. Can also identify landmarks and dominant colors in an online image to check the popularity of a product. Can moderate adult content and extract printed text messages but CANNOT translate text in an image

Application Insights

An Azure resource for query prediction telemetry. Can collect chat bot logs and includes powerful analytics tools to diagnose potential issues and process telemetry data with KQL (kusto query language)

Object Detection

An ML model and custom vision service that is the process of detecting items, such as animals or humans, in an image and returning the bounding block coordinates for each it has detected. Slightly more granular that image classification as it identifies specific images within the image. Can be used to evaluate compliance with business safety regulations by looking at building images to identify fire extinguishers etc. Ex: can identify and tag a vehicle but NOT assess level of damage.

Forecasting AI Workload

An ML model that analyzes patters in the provided data, and associates historical patterns to predict the possible outcome. EX: predicting whether an airplane arrives on time/early based off of historical traffic volumes and weather conditions etc.

Named Entity Recognition (NER)

An ML model that has the ability to identify various entities in text and classify them into predetermined types or categories, such as people, products, locations, etc. CANNOT be applied to images because it is a text analytics AI solution.

Hyperparameter

An ML tool that are adjustable parameters that let you control the model training/learning process/algorithm. This value is set BEFORE the leaning begins. Used to tune a machine learning model. Ex: # of runs or the sampling method

Turn

An action when creating the Azure bot framework. This action handles the received message and sends a message back to the user. You can add additional ____ when you want to handle different messages.

QnA Maker

An azure resource that provides authoring and query prediction. Uses a knowledge base with the pairs to answer a chatbot user's questions.

Fairness

a responsible AI principle that is concerned with treating people equally. An AI model should be interpreted to quantify the extent of how data influences a models prediction to help eliminate BIAS. EX: must operate without giving a certain advantage, cannot withhold information from, or allocating resources to a specific group of people.

Accountability

a responsible AI principle, that requires that the people involved in designing and developing AI base systems operate within a governance framework and follow defined ethical policies and legal standards

Azure Event HUb

a service for events processing and ingestion, CANNOT deploy ML here. (pre ingestion)

Azure Bot Framework SDK

is required to develop bots using code. Allows developers to send and receive messages with users on the configured channels. Supports C# java typescript and python. Can be integrated with LUIS (can add when you create it or later), QnA Maker (dispatch tool) as well as Power Virtual Agents (dispatch tool)

Semantic Segmentation

is the task of classifying each pixel in an image from a predefined set of classes or tagged object. EX: the pixels belonging to the car are classified in the class "car", the pixels corresponding to the street are labeled as "street", etc.