Copyright (c) 2025 MindMesh Academy. All rights reserved. This content is proprietary and may not be reproduced or distributed without permission.

Phase 7: Glossary

  • Accuracy: A performance metric for classification models that measures the percentage of predictions that were correct overall.
  • Algorithm: A procedure or set of rules that a machine learning system uses to learn patterns from data. It is the recipe used to create a model.
  • Amazon Aurora: A relational database service on AWS. When used with the pgvector extension, it can function as a vector database for RAG applications.
  • Amazon Bedrock: A fully managed AWS service that provides access to a variety of high-performing Foundation Models from leading AI companies via a single API, simplifying the development of generative AI applications.
  • Amazon Comprehend: A managed AWS AI service for Natural Language Processing (NLP) that uses machine learning to find insights and relationships in text, such as sentiment, entities, and key phrases.
  • Amazon Forecast: A managed AWS AI service that uses machine learning to deliver highly accurate time-series forecasts.
  • Amazon Fraud Detector: A managed AWS AI service that helps identify potentially fraudulent online activities, such as online payment fraud and the creation of fake accounts.
  • Amazon Lex: A managed AWS AI service for building conversational interfaces (chatbots) into any application using voice and text.
  • Amazon OpenSearch Service: A managed service for search and analytics. It includes a k-Nearest Neighbor (k-NN) search capability, making it a powerful and scalable option for vector databases.
  • Amazon Personalize: A managed AWS AI service that allows developers to build applications with the same machine learning technology used by Amazon.com for real-time personalized recommendations.
  • Amazon Polly: A managed AWS AI service that turns text into lifelike speech (Text-to-Speech, or TTS).
  • Amazon Q: An AI-powered assistant for work that can be tailored to a business, connecting to company data and systems to answer questions, summarize information, and perform tasks.
  • Amazon Rekognition: A managed AWS AI service that makes it easy to add image and video analysis to applications, detecting objects, people, text, and inappropriate content.
  • Amazon SageMaker: A fully managed service that provides every developer and data scientist with the ability to build, train, and deploy machine learning (ML) models at scale. It is the core "workbench" for custom ML development on AWS.
  • Amazon SageMaker Clarify: A feature of SageMaker that helps improve ML models by detecting potential bias in data and explaining how models make predictions.
  • Amazon SageMaker Data Wrangler: A feature of SageMaker that helps you aggregate and prepare data for machine learning with a visual, low-code interface.
  • Amazon SageMaker JumpStart: A feature of SageMaker that provides access to a wide range of publicly available, open-source Foundation Models, offering one-click deployment for fine-tuning and inference.
  • Amazon SageMaker Model Cards: A feature of SageMaker that provides a standardized way to document critical information about an ML model, acting as a "nutrition label" for transparency and governance.
  • Amazon SageMaker Model Monitor: A feature of SageMaker that automatically detects data drift and model quality degradation in deployed models.
  • Amazon SageMaker Pipelines: A service to build, automate, and manage end-to-end ML workflows (CI/CD for ML).
  • Amazon Textract: A managed AWS AI service that automatically extracts text, handwriting, and data from scanned documents.
  • Amazon Transcribe: A managed AWS AI service that makes it easy for developers to add speech-to-text (ASR) capability to their applications.
  • Amazon Translate: A managed AWS AI service that delivers fast, high-quality, and affordable language translation.
  • Artificial Intelligence (AI): The broad scientific field of creating machines or software that can perform tasks that typically require human intelligence.
  • AWS Artifact: A service that provides on-demand access to AWS's security and compliance reports, such as ISO certifications and SOC reports.
  • AWS Audit Manager: A service that helps you continuously audit your AWS usage to simplify how you assess risk and compliance with regulations and industry standards.
  • AWS CloudTrail: An AWS service that logs every API call made in your account, providing a detailed audit trail of actions taken by users, roles, or AWS services.
  • AWS Config: An AWS service that assesses, audits, and evaluates the configurations of your AWS resources, helping you ensure compliance with policies.
  • AWS PrivateLink: A networking service that provides private, secure connectivity between your VPCs and AWS services, without exposing your traffic to the public internet.
  • Bias: A systematic error in an AI system that results in unfair or discriminatory outcomes, often stemming from unrepresentative training data.
  • BLEU (Bilingual Evaluation Understudy): An automated metric used to evaluate the quality of text that has been machine-translated from one language to another by comparing it to high-quality human translations.
  • Classification: A type of supervised learning that predicts a categorical label (e.g., "spam" or "not spam," "dog" or "cat").
  • Clustering: A type of unsupervised learning that groups data points based on their similarities, without any predefined labels.
  • Context Window: The maximum amount of text (input prompt + output generation) that a Foundation Model can process at one time.
  • Custom Model: A machine learning model that is trained by an organization on its own private, specific dataset to solve a problem unique to its business.
  • Data Lineage: The practice of tracking the origin, movement, and transformation of data throughout its lifecycle. Crucial for reproducibility and auditing.
  • Data Residency: The requirement that certain data must be physically stored within a specific geographical location to comply with local laws and regulations.
  • Data Retention: A policy that defines how long data must be kept for legal, regulatory, or business reasons.
  • Deep Learning (DL): A subfield of Machine Learning that uses complex, multi-layered neural networks to learn from vast amounts of data, especially powerful for unstructured data like images and text.
  • Diffusion Model: A type of generative model, primarily used for creating high-quality images, that works by starting with random noise and gradually refining it into a coherent image based on a text prompt.
  • Embeddings: A numerical representation (a vector or list of numbers) of text, images, or other data that captures its semantic meaning. Similar concepts have similar embedding vectors.
  • Encryption: The process of encoding data to protect it from unauthorized access. Can be "at rest" (when stored) or "in transit" (when moving over a network).
  • Exploratory Data Analysis (EDA): The initial process of analyzing a dataset to understand its main characteristics, discover patterns, and spot anomalies.
  • Fairness: The ethical goal of ensuring that an AI model's predictions do not create discriminatory or unjust outcomes for different demographic groups.
  • Feature Engineering: The process of selecting, transforming, and creating input variables ("features") from raw data to improve the performance of a machine learning model.
  • Few-shot Prompting: A prompt engineering technique where multiple examples (2-5) of the desired task are provided within the prompt to guide the model's response.
  • Fine-tuning: The process of taking a general-purpose, pre-trained Foundation Model and continuing to train it on a smaller, high-quality, task-specific dataset to adapt it for a particular purpose.
  • Foundation Model: A very large, general-purpose model trained on a massive amount of broad data, which can be adapted ("fine-tuned") for a wide range of downstream tasks.
  • Generative AI: A class of artificial intelligence that can create new, original content, such as text, images, music, or code, rather than just predicting a label or a number.
  • Guardrails for Amazon Bedrock: A safety feature that allows you to implement policies to control the types of content your generative AI application will generate, by defining denied topics and content filters.
  • Hallucination: A phenomenon where a generative AI model produces text that is plausible and well-written but is factually incorrect, invented, or nonsensical.
  • Hyperparameter Tuning: The process of finding the optimal settings (hyperparameters) for a machine learning algorithm to maximize the performance of the resulting model.
  • IAM (Identity and Access Management): The foundational AWS service for securely controlling access to AWS services and resources by managing users, groups, and permissions.
  • Inferencing (or Prediction): The process of using a trained machine learning model to make a prediction on new, unseen data.
  • Intellectual Property (IP) Infringement: A legal risk associated with generative AI where the model's output is substantially similar to copyrighted material it was trained on.
  • Labeled Data: Data where each example is tagged with the correct answer or "label." It is the required input for supervised learning.
  • Large Language Model (LLM): A type of Foundation Model trained specifically on vast quantities of text data, designed to understand, process, and generate human-like language.
  • Latency: The time delay between a request being sent to a model and a response being received. Low latency is critical for real-time applications.
  • Machine Learning (ML): A subfield of AI where systems learn patterns directly from data, rather than being explicitly programmed with rules, to make predictions or decisions.
  • MLOps (Machine Learning Operations): A set of practices that combines machine learning, DevOps, and data engineering to automate and streamline the ML lifecycle from development to production.
  • Model: The output artifact of the machine learning training process. The model contains the learned patterns and is used to make inferences.
  • Multi-modal: An AI model that can process and understand information from multiple types of data (modalities), such as text, images, and audio, simultaneously.
  • Nondeterminism: A characteristic of generative AI models where asking the same prompt multiple times can produce different, non-identical responses.
  • Opaque Model (Black-box): A complex model, such as a deep neural network, whose internal decision-making process is not easily understood by humans.
  • On-Demand (Pricing): A pay-per-use pricing model for AI services where you are billed based on the amount of data processed, typically measured in tokens.
  • pgvector: An open-source extension for PostgreSQL databases (including Amazon RDS and Aurora) that enables the storage and searching of vector embeddings.
  • Pre-trained Model: A model that has been trained on a large, general dataset by a third party and can be used directly for common tasks or as a starting point for fine-tuning.
  • Prompt Engineering: The art and science of designing effective inputs (prompts) to guide a generative AI model to produce the desired output.
  • Provisioned Throughput: A pricing model for AI services where you purchase dedicated capacity for a fixed price to ensure consistent performance and potentially lower costs for high-volume, stable workloads.
  • RAG (Retrieval Augmented Generation): A design pattern that enhances a Foundation Model's reliability by retrieving relevant information from an external knowledge base and adding it to the prompt. This grounds the model in factual data, reducing hallucinations and enabling it to answer questions about private information.
  • Regression: A type of supervised learning that predicts a continuous numerical value (e.g., a price, a temperature, a sales forecast).
  • Reinforcement Learning (RL): A type of machine learning where an "agent" learns to make decisions by interacting with an environment and receiving rewards or penalties for its actions.
  • Reinforcement Learning from Human Feedback (RLHF): An advanced fine-tuning technique that trains models to align with human preferences by having humans rank and compare different model outputs.
  • Responsible AI: A comprehensive approach to developing and deploying AI systems that prioritizes fairness, inclusivity, robustness, safety, and transparency to maximize benefits while minimizing risks.
  • ROUGE (Recall-Oriented Understudy for Gisting Evaluation): An automated metric used to evaluate the quality of a machine-generated summary by measuring its word overlap with a human-written reference summary.
  • Shared Responsibility Model: The AWS security framework where AWS is responsible for the security OF the cloud (infrastructure, managed services), and the customer is responsible for security IN the cloud (data, access configuration, responsible AI).
  • Single-shot Prompting: A prompt engineering technique where a single example of the desired task is provided within the prompt to guide the model's response.
  • Structured Data: Highly organized data that adheres to a predefined format, typically in rows and columns, like in a spreadsheet or database.
  • Supervised Learning: A type of machine learning where the model is trained on labeled data, learning to map input features to the correct output label.
  • Token: A common sequence of characters that an LLM uses to process text. A token can be a whole word, a part of a word, or punctuation.
  • Transparent Model (White-box): A model, such as a decision tree, whose internal decision-making process is easily understood by humans.
  • Unlabeled Data: Raw data with no predefined answers or labels. It is the required input for unsupervised learning.
  • Unstructured Data: Data that does not have a predefined organizational structure, such as the text of an email, an audio file, a video, or an image.
  • Unsupervised Learning: A type of machine learning where the model is given unlabeled data and tasked with finding hidden patterns or structures on its own.
  • Vector Database: A specialized database designed to efficiently store and search on embeddings using similarity search. It is the foundational technology for the "Retrieve" step in RAG.
  • Vector Search (Similarity Search): The process of finding vectors in a database that are mathematically closest (most similar) to a given query vector, enabling search based on semantic meaning rather than keywords.
  • Zero-shot Prompting: A prompt engineering technique where the model is asked to perform a task without being given any examples in the prompt, relying solely on its pre-trained knowledge.