Microsoft Azure AI Fundamentals (AI-900) flashcards

Inclusiveness in Responsible AI means that AI systems should empower everyone and engage diverse human experiences and perspectives.

• Aims to design AI that is accessible and usable by people with varying abilities, backgrounds, and needs.
• Involves considering diverse user groups during design, development, and testing.
• Helps prevent exclusion and ensures AI benefits a broader population.

Memory trick: Inclusiveness: AI includes everyone, like a big group hug.

Using AI to enable computers to understand and interpret visual information from the world, such as images and videos.

• Key tasks include object detection, image classification, and semantic segmentation.
• Crucial for pattern recognition in visual data.
• Applications range from medical imaging to autonomous vehicles.

Memory trick: Vision AI sees the forest, not just the trees.

The principle that AI systems should perform reliably, consistently, and safely, minimizing the risk of unintended harm.

• Crucial for systems interacting with the physical world or critical infrastructure.
• Requires robust testing, validation, and monitoring.
• Aims to prevent errors, failures, and unintended negative consequences.

Memory trick: AI must be RELIABLE and SAFE, like a trusted guardian.

Ensuring AI systems treat all people fairly and equitably, without perpetuating or amplifying societal biases.

• Aims to prevent discrimination based on race, gender, age, etc.
• Requires diverse and representative training data.
• Involves regular auditing for biased outcomes.

Memory trick: AI must be FAIR to everyone, like a good judge.

Ensuring AI systems apply content moderation policies consistently and equitably across all users and content, without bias or disproportionate impact.

• Avoids censoring legitimate speech from specific groups.
• Prevents harmful content targeting certain communities.
• Requires diverse and unbiased training data.

Memory trick: Fair AI, equal voice, no biased choice.

An AI workload focused on creating new, original content like text, images, audio, or video, often based on learned patterns from existing data.

• Used for content creation, design, code generation.
• Can produce realistic and novel outputs.
• Aims to mimic human creativity.

Memory trick: GENERATING new ideas is Generative AI's superpower.

A branch of AI that enables computers to understand, interpret, and generate human language.

• Focuses on text and speech data.
• Used in chatbots, sentiment analysis, machine translation.
• Essential for human-computer interaction through language.

Memory trick: AI understands human words, not just pictures or numbers.

The principle that AI systems should be understandable, allowing users and stakeholders to comprehend their decision-making processes.

• Aims for explainability and interpretability of AI models.
• Helps build trust and allows for debugging/improvement.
• Crucial for critical applications where 'why' matters.

Memory trick: AI should be TRANSPARENT, like clear glass, so you see through it.

Reliability and Safety in Responsible AI means that AI systems should perform consistently, dependably, and safely, and should be designed to minimize unintended harm.

• Crucial for high-stakes applications where errors can have severe consequences (e.g., healthcare, autonomous systems).
• Involves robust testing, validation, and mechanisms for graceful degradation or human override.
• Aims to build trust by ensuring AI systems are trustworthy and do not pose risks.

Memory trick: Reliability and Safety: AI must be reliable and safe, like a seatbelt.

An AI workload that combines AI services (like NLP and computer vision) with search technologies to extract insights, relationships, and structured information from large volumes of unstructured data.

• Transforms unstructured data into actionable knowledge.
• Often used for document understanding, sentiment analysis, and content search.
• Leverages various AI services to achieve its goal.

Memory trick: Mining Knowledge Makes Meaningful Insights.

Ensuring AI systems protect personal data, respect individual privacy, and are secure against vulnerabilities.

• Involves data minimization, anonymization, and robust security measures.
• Aims to prevent unauthorized access, use, or disclosure of sensitive information.
• Crucial for maintaining public trust in AI applications.

Memory trick: AI must guard your SECRETS and keep you SAFE.

An AI workload focused on identifying rare events, observations, or patterns that differ significantly from the majority of the data.

• Used for fraud detection, equipment failure prediction, cybersecurity.
• Identifies 'outliers' or 'novelties' in data.
• Crucial for maintaining system integrity and security.

Memory trick: Finding the 'odd one out' is Anomaly Detection's job.

The principle that people should be accountable for the design, development, deployment, and impact of AI systems.

• Ensures human oversight and responsibility for AI actions.
• Addresses who is liable when AI systems cause harm or make errors.
• Critical for building trust and managing risks associated with AI.

Memory trick: Someone must be ACCOUNTABLE, like the captain of a ship.

A subfield of computational linguistics that translates text or speech from one natural language into another, typically using AI techniques.

• Enables cross-lingual communication.
• Relies heavily on Natural Language Processing (NLP).
• Can be statistical, rule-based, or neural network-based.

Memory trick: Language barriers fall with NLP's call.

Ensuring the logic, factors, and processes behind AI-driven pricing decisions are understandable, explainable, and auditable.

• Allows for justification of price changes.
• Facilitates regulatory compliance and prevents exploitation.
• Builds trust with consumers by demystifying pricing.

Memory trick: Transparent pricing, clear and advising.

An AI workload focused on enabling computers to 'see,' interpret, and understand the visual world from images and videos.

• Used for object detection, facial recognition, medical image analysis.
• Processes pixels and visual patterns.
• Key for automation in visual inspection and surveillance.

Memory trick: Computers using their EYES are doing Computer Vision.

The use of data, statistical algorithms, and machine learning techniques to identify the likelihood of future outcomes based on historical data.

• Forecasts future events and behaviors.
• Often used for optimization and decision support.
• Relies on patterns and relationships in data.

Memory trick: Predictive analytics, future paths it tracks.

Privacy and Security in Responsible AI ensures that AI systems protect personal data, respect individual privacy, and are secure from cyber threats and unauthorized access.

• Involves data anonymization, encryption, and access controls.
• Aims to prevent data breaches and misuse of sensitive information.
• Crucial for maintaining public trust and compliance with regulations.

Memory trick: P.S. I love responsible AI: Privacy and Security first!

A phenomenon where a machine learning model learns the training data too precisely, including noise and specific patterns, leading to excellent performance on training data but poor generalization to new, unseen data.

• High training accuracy, low validation/test accuracy.
• Caused by overly complex models or insufficient training data.
• Mitigation techniques include regularization, early stopping, cross-validation, more data, feature selection.

Memory trick: Overfitting is 'Over-complicated' and 'Over-confident' on training data.

A supervised machine learning task that involves predicting a continuous numerical value.

• Output is a number (e.g., price, temperature, sales quantity).
• Requires labeled training data.
• Common algorithms include Linear Regression, Decision Tree Regressor, Support Vector Regressor.

Memory trick: Regression predicts 'Real' numbers, like sales figures.

Collaborative filtering is a technique used by recommendation systems that makes predictions about a user's interests by collecting preferences from many users. It identifies users with similar tastes and recommends items liked by those 'similar' users.

• Relies on user-item interaction data (ratings, purchases, views).
• Two main types: user-based and item-based.
• Suffers from 'cold start' problem for new users/items.

Memory trick: Collaborative: Users 'collaborate' to help each other discover new things.

A data preprocessing technique used to standardize or normalize the range of independent variables (features) in a dataset.

• Prevents features with larger magnitudes from dominating distance-based algorithms.
• Common methods: Min-Max Scaling (Normalization) and Standardization (Z-score normalization).
• Essential for algorithms like K-Nearest Neighbors, Support Vector Machines, and neural networks.

Memory trick: Feature Scaling 'levels' the 'playing field' for numerical data.

A dataset where the number of observations for one class (or category) is significantly lower than for other classes.

• Can lead to misleading accuracy metrics.
• Models may prioritize the majority class.
• Requires specialized techniques (e.g., oversampling, undersampling, using different metrics like F1-score, precision, recall) to address.

Memory trick: Imbalanced data 'skews' the 'score' and makes it 'unbalanced'.

A type of machine learning that finds patterns or structures in unlabeled data, without human intervention.

• Works with data that has no predefined output labels.
• Common tasks include clustering, dimensionality reduction, anomaly detection.
• Used for exploratory data analysis, customer segmentation, data compression.

Memory trick: Learning: With a Teacher or Discovery Alone

A type of machine learning that finds hidden patterns or data groupings in unlabeled datasets, with clustering being a primary technique.

• No labeled output variable is provided during training.
• Aims to discover structures and relationships within data.
• Clustering algorithms group similar data points together.

Memory trick: Unsupervised finds 'unseen' groups in purchase data.

Clustering is an unsupervised machine learning technique used to group a set of objects in such a way that objects in the same group (called a cluster) are more similar to each other than to those in other groups. It is used to discover hidden patterns or natural groupings in data.

• An unsupervised learning method.
• Does not require labeled data.
• Identifies intrinsic groupings within data.
• Common algorithms include K-Means, DBSCAN, Hierarchical Clustering.

Memory trick: Unsupervised learning 'clusters' and 'reduces' without a teacher.

Image segmentation is a computer vision technique that divides an image into multiple segments or regions, often down to the pixel level. Each segment corresponds to an object or a distinct part of the image, allowing for precise localization and boundary detection.

• Assigns a label to every pixel in an image.
• Provides pixel-level understanding of an image.
• Used for precise object boundary detection.
• Applications include medical imaging, autonomous driving, satellite imagery analysis.

Memory trick: Vision tasks range from 'classifying' a whole picture to 'segmenting' every pixel.

Recall, also known as sensitivity or the true positive rate, is a metric that measures the proportion of actual positive cases that were correctly identified by the model. It is particularly important when the cost of false negatives is high.

• Calculated as True Positives / (True Positives + False Negatives).
• Focuses on minimizing false negatives.
• High recall means fewer actual positive cases are missed.
• Crucial in medical diagnosis, fraud detection, and safety systems.

Memory trick: Remembering 'PR-FACT': Precision, Recall, F1, Accuracy, Confusion Matrix, True/False.

A technique used to convert categorical variables into a numerical format that machine learning algorithms can understand, by creating new binary features for each category.

• Creates a new binary column for each unique category.
• Prevents the model from assuming an ordinal relationship between categories.
• Commonly used for nominal (unordered) categorical data.

Memory trick: Transforming names into numbers, carefully.

A machine learning technique that identifies data points that do not conform to an expected pattern in a dataset, without requiring labeled examples of anomalies for training.

• Used when 'normal' behavior is known, but 'anomalous' behavior is not or is rare.
• Often involves models that learn the distribution of normal data.
• Outliers are flagged based on deviation from the learned normal distribution.

Memory trick: Unsupervised 'finds' the 'unseen' problems in the machine data.

Azure Machine Learning is a cloud-based service for accelerating and managing the machine learning project lifecycle. It empowers developers and data scientists with a wide range of tools and services to build, train, deploy, and manage machine learning models.

• End-to-end ML platform (data prep, training, deployment).
• Supports various ML frameworks (PyTorch, TensorFlow, scikit-learn).
• Offers MLOps capabilities for model management and monitoring.
• Enables deployment to various targets (web service, IoT Edge).

Memory trick: Azure ML is the 'hub' for all machine learning tasks, from creation to deployment.

A supervised machine learning task where the model learns to predict a discrete category or class label for new input data.

• Output is a discrete value (e.g., 'spam'/'not spam', 'cat'/'dog', 'yes'/'no').
• Requires labeled training data.
• Common algorithms include Logistic Regression, Support Vector Machines, Decision Trees.

Memory trick: Learning with a teacher to hit the bullseye.

A False Negative occurs when a machine learning model incorrectly predicts a negative outcome when the actual outcome is positive. It's a 'missed opportunity' or a 'missed detection'.

• Model predicts 'No', but the truth is 'Yes'.
• Can have severe consequences depending on the application (e.g., missing a disease, missing fraud).
• Often minimized by tuning models for higher recall.

Memory trick: A 'False Negative' is a 'No' when it should have been a 'Yes' – a truly missed opportunity.

A regression metric that measures the average of the absolute differences between predictions and actual observations. It quantifies the average magnitude of errors, without considering their direction.

• Calculated as the average of |actual - predicted|.
• Robust to outliers compared to Mean Squared Error (MSE).
• Units are the same as the target variable, making it easily interpretable.

Memory trick: Measuring how close our predicted line is to the real data points.

Supervised learning is a machine learning paradigm where an algorithm learns from a labeled dataset. It builds a model that maps input features to an output label, which can then be used to predict outcomes for new, unseen data.

• Requires labeled training data.
• Tasks include classification and regression.
• Aims to predict an output variable based on input variables.
• Learns from examples with known correct answers.

Memory trick: Machines learn in three main 'modes' — with a teacher, on their own, or by trying.

A graph showing the performance of a classification model at all classification thresholds. It plots two parameters: True Positive Rate (TPR) on the y-axis and False Positive Rate (FPR) on the x-axis.

• TPR is also known as Sensitivity or Recall.
• FPR is calculated as 1 - Specificity.
• The Area Under the Curve (AUC-ROC) provides a single metric for overall performance.
• Useful for evaluating models on both balanced and imbalanced datasets.

Memory trick: Drawing lines to see how good our predictions are.

A phenomenon in machine learning where a high overall accuracy score can be misleading, particularly in datasets with highly imbalanced classes. A model can achieve high accuracy by simply predicting the majority class, while performing poorly on the minority class.

• Occurs in datasets where one class significantly outnumbers others.
• A simple 'always predict majority' model can yield high accuracy.
• Requires other metrics like Precision, Recall, F1-score, or AUC-ROC for proper evaluation.

Memory trick: When the scales are tipped, accuracy can lie.

The ratio of correctly predicted positive observations to the total predicted positive observations. It answers: 'Of all the times we predicted positive, how many were actually positive?'

• Focuses on minimizing False Positives.
• Important when the cost of a false positive is high.
• Calculated as True Positives / (True Positives + False Positives).

Memory trick: Every Metric Tells a Tale of Model Performance

The process of selecting a subset of relevant features (variables, predictors) for use in model construction.

• Improves model performance by reducing overfitting.
• Reduces training time and computational cost.
• Enhances model interpretability.

Memory trick: Clean Data, Clear Decisions

Area Under the Receiver Operating Characteristic Curve; an evaluation metric for binary classification models that measures the model's ability to distinguish between classes across all possible thresholds.

• Ranges from 0 to 1; higher is better.
• 1.0 indicates perfect classification, 0.5 indicates random guessing.
• Robust to imbalanced datasets.
• Considers both true positive rate (sensitivity) and false positive rate (1-specificity).

Memory trick: AUC-ROC curves show 'how much' a model 'Rocks' at distinguishing.

Microsoft's object storage solution for the cloud. Blob storage is optimized for storing massive amounts of unstructured data.

• Stores unstructured data (text, binary, images, video).
• Highly scalable and durable.
• Accessible via REST APIs, SDKs, and Azure portal.

Memory trick: Azure Storage: Pick the Right Bin for Your Data

The Receiver Operating Characteristic (ROC) curve is a graphical plot that illustrates the diagnostic ability of a binary classifier system as its discrimination threshold is varied. It plots the True Positive Rate (Sensitivity) against the False Positive Rate (1-Specificity) at various thresholds.

• Evaluates binary classification models.
• Shows performance across all possible thresholds.
• Plots True Positive Rate vs. False Positive Rate.
• Area Under the Curve (AUC-ROC) summarizes overall performance.

Memory trick: Visualizing performance: 'ROC' for thresholds, 'Confusion' for counts.

A machine learning paradigm where an agent learns to make optimal decisions by interacting with an environment, receiving rewards for desired actions and penalties for undesired ones, aiming to maximize cumulative reward.

• Involves an agent, environment, actions, states, and rewards.
• Learns through trial and error.
• Often used for control problems, robotics, game playing, autonomous systems.

Memory trick: Reinforcement 'Rewards' the agent for 'Right' driving decisions.

An Azure AI service feature used to train a custom image classification model that can categorize images into specific, user-defined classes based on provided training data.

• You provide the images and define the labels (classes).
• Ideal for niche classification tasks not covered by generic pre-trained models.
• Can be used to classify entire images or identify regions within images (object detection).

Memory trick: Custom Vision recognizes unique images with custom training.

A computer vision technique that assigns a class label to every pixel in an image, effectively segmenting the image into regions corresponding to different object categories.

• Provides pixel-level understanding of an image.
• Useful for precise area calculation, medical imaging, autonomous driving.
• Distinguishes between different classes of objects, but not individual instances within the same class.

Memory trick: Segmentation carves images into meaningful pieces.

Using object detection models to identify and localize individual instances of objects within an image, thereby enabling accurate counting of those objects.

• Each detected object gets a bounding box and a class label.
• Useful for inventory, traffic analysis, and wildlife monitoring.
• Provides both 'what' and 'where' information.

Memory trick: Detect objects to count and place them.

Combining Speech-to-Text with Text Analytics allows for the conversion of spoken audio into text, followed by the extraction of insights like sentiment, keywords, and entities from the transcribed content.

• Common in call center analytics, voice assistants, meeting transcription.
• Azure Speech Service provides the transcription part.
• Azure Text Analytics provides the NLP analysis part.

Memory trick: Speak, Transcribe, Analyze, Understand.

An Azure AI service that allows developers to build, deploy, and improve custom image classification and object detection models for specific use cases.

• Requires training with your own labeled images.
• Ideal for niche or specialized computer vision tasks not covered by general pre-trained models.
• Supports both classification (what is in the image) and object detection (where are specific things in the image).

Memory trick: Custom Vision: Train your eye for unique targets.

Azure Translator is a cloud-based neural machine translation service that enables fast and accurate language translation.

• Supports over 100 languages and dialects.
• Used for real-time translation, document translation, and website localization.
• Can be integrated into applications, websites, and workflows.

Memory trick: Translator Bridges Language Gaps Swiftly.

An NLP capability that allows users to train models to identify and extract domain-specific or organization-specific entities from text.

• Extends standard NER for unique terms.
• Requires labeled training data.
• Useful for specialized industries (medical, legal, finance).

Memory trick: Custom NER for custom names.

Optical Character Recognition (OCR) is a technology that enables conversion of images of text (typewritten, handwritten or printed) into machine-encoded text.

• Essential for digitizing physical documents.
• Allows text searchability and editing of scanned content.
• Often a prerequisite for further NLP tasks on scanned documents.

Memory trick: OCR Makes Image Text Readable.

An Azure AI service that uses machine learning to identify and extract key-value pairs, text, and tables from documents.

• Specializes in structured data extraction from forms and documents.
• Can handle both standardized and non-standardized document layouts.
• Offers pre-built models and custom model training.

Memory trick: Forms Recognize Data from Documents.

An advanced NLP technique that identifies specific aspects or entities in text and then determines the sentiment expressed towards each of those individual aspects.

• Provides fine-grained sentiment analysis.
• Identifies target entities (e.g., product features).
• Crucial for detailed product feedback analysis.

Memory trick: ABSA Analyzes By Specific Aspects

An Azure Cognitive Service for Language component that enables applications to understand natural language input by identifying user intent and extracting relevant entities.

• Core component for conversational AI (chatbots, virtual assistants).
• Focuses on intent and entity recognition.
• Requires training with example utterances.

Memory trick: LUIS helps your app understand what you 'mean'.

The process of identifying and filtering inappropriate or policy-violating content across multiple languages.

• Often requires Machine Translation as a precursor.
• Utilizes NLP for sentiment, toxicity, and policy checks.
• Essential for global communication platforms.

Memory trick: Translate to understand, then moderate to comply.

The NLP task of assigning one or more predefined categories or labels to a piece of text.

• Used for spam detection, content routing, sentiment analysis (as a binary classification).
• Requires a set of predefined categories.
• Can be custom-trained for specific domains.

Memory trick: Classify text to put it in the right folder.

A technology capable of identifying or verifying a person from a digital image or a video frame.

• Compares detected faces against a database of known faces.
• Requires a 'face print' or 'template' for each individual.
• Used for security, access control, and personalized experiences.

Memory trick: To recognize a face, you need Face Recognition.

An Azure AI Computer Vision capability that generates a natural language description (caption) of the content within an image.

• Provides a concise summary of the visual content.
• Useful for accessibility, content indexing, and search.
• Differs from tagging, which provides keywords, and object detection, which identifies specific items.

Memory trick: Images can be tagged, detected, or fully described.

A combination of NLP tasks to understand the emotional tone of text and extract its core topics or concepts.

• Commonly used in customer feedback analysis.
• Sentiment provides emotional context.
• Key phrases reveal discussed subjects.

Memory trick: Feel the sentiment, find the key phrases.

Object detection identifies and locates multiple objects within an image, while image classification assigns a single label to the entire image.

• Classification: 'This image contains a dog.'
• Detection: 'There is a dog at [x,y,w,h] and another dog at [x2,y2,w2,h2].'
• Both are fundamental computer vision tasks.

Memory trick: To count individuals, you must detect them first.