Microsoft Azure AI Fundamentals (AI-900)

The principle that AI systems should treat all people fairly and equitably, without discriminating against individuals or groups based on characteristics like race, gender, or socioeconomic status.

• Aims to prevent bias and discrimination.
• Requires careful consideration of training data.
• Ensures equitable outcomes for diverse user groups.

Ensuring AI systems are understandable, and their decision-making processes and rationale are explainable.

• Crucial for trust, auditing, and debugging complex AI models.
• Involves techniques like Explainable AI (XAI).
• Allows human experts to validate AI's reasoning.

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.

A field of AI that enables computers to 'see', interpret, and understand the visual world.

• Processes images and videos.
• Used for object detection, facial recognition, defect inspection.
• Involves tasks like classification, segmentation, and tracking.

Accountability in Responsible AI means that people should be ultimately responsible for AI systems and their impact, with clear mechanisms for oversight and redress.

• Ensures human oversight and ultimate decision-making, especially in high-stakes scenarios.
• Establishes clear lines of responsibility for AI system development, deployment, and operation.
• Crucial for building trust and addressing potential harms caused by AI.

A type of AI that can create new content, such as text, images, audio, or video, that resembles real-world data.

• Learns patterns from existing data to produce novel outputs.
• Often uses large language models (LLMs) for text generation.
• Applications include content creation, design, and data augmentation.

Anomaly Detection is an AI workload focused on identifying unusual data points, events, or observations that deviate significantly from the majority of the data.

• Used in fraud detection, cybersecurity, predictive maintenance, and medical diagnosis.
• Relies on learning 'normal' behavior to flag 'abnormal' instances.
• Can identify rare events that might indicate a problem or opportunity.

AI systems should perform reliably, consistently, and safely, and be able to respond to unexpected conditions or changes.

• Crucial for mission-critical and safety-critical applications.
• Involves robustness, resilience, and error handling.
• Aims to prevent unintended harm to people or property.

The principle that AI systems should protect personal data, respect privacy, and be secure against unauthorized access and malicious attacks.

• Crucial when handling sensitive personally identifiable information (PII).
• Involves data encryption, access controls, and robust cybersecurity measures.
• Aims to prevent data breaches and ensure data confidentiality.

An AI workload focused on enabling computers to understand, interpret, and generate human language.

• Used for sentiment analysis, chatbots, language translation.
• Processes text and spoken language.
• Key for human-computer interaction through language.

Knowledge Mining is an AI workload that uses AI to extract information, discover patterns, and gain insights from large volumes of unstructured and semi-structured data.

• Transforms unstructured data into structured, searchable information.
• Often involves techniques like natural language processing, computer vision, and search.
• Helps organizations find hidden insights and automate information retrieval.

AI systems should empower everyone and engage people, designing for diverse human experiences and abilities.

• Aims to make AI accessible to a wide range of users.
• Considers diverse languages, accents, abilities, and backgrounds.
• Prevents exclusion by design.

A branch of AI that uses statistical algorithms and machine learning techniques to identify the likelihood of future outcomes based on historical and current data.

• Forecasts future events or behaviors.
• Utilizes historical data, statistical modeling, and machine learning.
• Commonly applied in fraud detection, risk assessment, and maintenance prediction.

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.

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.

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.

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.

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.

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.

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.

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.

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.

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.

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.