AWS Certified AI Practitioner Foundational (AIF-C01) Study Guide

Comprehensive vocabulary flashcards covering key concepts, services, evaluation metrics, and security mechanisms for the AWS Certified AI Practitioner Foundational (AIF-C01) exam.

Prompt Engineering

The process where you guide generative artificial intelligence (generative AI) solutions to generate desired outputs.

Artificial Intelligence

A field of computer science dedicated to solving cognitive problems commonly associated with human intelligence.

Conversational AI

A technology that makes software capable of understanding and responding to voice-based or text-based human conversations.

Generative AI

A type of AI that can create new content and ideas, including conversations, stories, images, videos, and music.

Embedding

A numerical representation of real-world objects that machine learning (ML) and artificial intelligence (AI) systems use to understand complex knowledge domains like humans do.

Supervised Learning

A prominent type of machine learning requiring a supervisor, used across a vast range of applications.

Classification

A supervised learning technique that assigns labels or categories to new, previously unseen data examples using a learned model.

Regression

A supervised learning technique that predicts continuous or numerical values given one or more input variables.

Unsupervised Learning

A type of machine learning where the algorithm tries to discover hidden patterns or structures within data without any prior information or guidance.

Clustering

An unsupervised method that divides data into clusters based on similar traits or distances between data points to better understand the characteristics of a particular cluster.

Dimensionality Reduction

An unsupervised learning strategy that minimizes the number of features or dimensions in a dataset while retaining the most relevant information or patterns.

Reinforcement Learning

A machine learning approach where an agent continuously improves its model by trial and error as it interacts with its surroundings and analyzes feedback from prior versions.

Intelligent Document Processing (IDP)

A software tool that extracts and categorizes information from unstructured or structured data, generates summaries, and delivers actionable insights.

Factors to Consider When Selecting a Generative AI Model

Five key evaluation criteria: Performance requirements, Constraints, Capabilities, Compliance, and Cost.

Amazon Q Business

A generative AI-powered assistant that can answer queries, generate content, provide summaries, and complete tasks based on data in your organization.

Responsible AI

Procedures and principles that ensure AI systems are transparent and trustworthy while minimizing potential risks and adverse effects.

Challenges of Generative AI

Key risks and concerns, including Toxicity, Hallucinations, Intellectual Property issues, Plagiarism and Cheating, and Disruptions of Nature Work.

Multimodal Models

Models designed to process inputs from various sources simultaneously, including text, images, audio, and video.

Amazon SageMaker

A fully managed AWS service that provides a complete machine learning lifecycle, including data preparation, model building, training, tuning, and deployment.

Natural Language Processing (NLP)

A field within artificial intelligence focused on the interaction between computers and human languages.

Amazon SageMaker Clarify

A machine learning tool that provides insights to improve model fairness and transparency by examining potential bias in datasets and model predictions.

Deep Learning

A technique in artificial intelligence that teaches computers to process data in a manner inspired by the human brain.

Neural Networks

An artificial intelligence technique that teaches computers to interpret data in a manner inspired by the human brain.

Computer Vision

An AI field enabling machines to recognize people, places, and things in photographs with human-level accuracy or greater, operating at higher speeds and efficiency.

Amazon SageMaker Feature Store

A fully managed, purpose-built repository for storing, sharing, and managing machine learning features.

Amazon SageMaker Model Monitor

A tool that automatically monitors machine learning models in production, applies criteria to detect drift, and alerts you when quality concerns arise.

Fine-Tuning

The process of training a previously trained model on a fresh dataset rather than starting from scratch, also known as transfer learning.

Amazon SageMaker JumpStart

A hub to rapidly evaluate, compare, and select foundation models (FMs) for tasks like article summarizing and image generation based on predefined quality and responsibility metrics.

Amazon Bedrock

A fully managed solution that makes high-performing foundation models from top AI startups and Amazon accessible through a single, common API.

PartyRock

An Amazon Bedrock playground providing a fun, hands-on environment where users can intuitively build generative AI applications in just a few steps.

Retrieval Augmented Generation (RAG)

A method that optimizes the output of a large language model (LLM) by referencing an external knowledge base containing company-specific or industry-specific data.

Negative Prompts

Prompts used in generative models to specify what content or features should be excluded from the generated output.

Model Latent Space

The abstract space where machine learning models map input data into feature representations used to generate outputs.

Reinforcement Learning from Human Feedback (RLHF)

A technique that incorporates human feedback to help machine learning models make predictions more efficiently and accurately while maximizing rewards.

ROUGE (Recall-Oriented Understudy for Gisting Evaluation)

A metric used to evaluate the quality of text summarization by calculating the overlap between generated summaries and reference summaries.

BLEU (Bilingual Evaluation Understudy)

A metric for assessing machine translation quality by comparing the similarity between machine output and human reference translations.

BERTScore

A metric for evaluating text generation models by assessing similarity at the token level using BERT embeddings.

Amazon Augmented AI (Amazon A2I)

A service that makes it easy to integrate human review workflows into machine learning predictions to ensure high-quality results.

Amazon SageMaker Model Cards

Documents providing centralized, key details about machine learning models, including performance metrics, intended use, and compliance information.

Amazon Macie

A fully managed data security and privacy service that applies machine learning to discover, classify, and protect sensitive data in AWS.

Prompt Injection

A security vulnerability where malicious input is embedded into prompts to manipulate or hijack the output of language models.

Generative AI Security Scoping Matrix

A framework used to identify, scope, and manage security risks associated with deploying and using generative AI models.

Amazon Transcribe

A fully managed speech-to-text service that automatically converts audio files into accurate, readable text.

Amazon Comprehend

A managed NLP service used to extract key phrases, sentiment, entities, and other key insights from unstructured text.

Amazon Lex

An AWS service used to build conversational chatbots and interfaces based on recognizable intents and entities.

Amazon Polly

A text-to-speech service that converts written content into lifelike human speech for voice-enabled applications.

Exploratory Data Analysis (EDA)

The process of analyzing datasets to summarize their primary characteristics, frequently utilizing visualizations to uncover trends, patterns, and anomalies.

Hyperparameter Tuning

The process of optimizing the external configuration parameters that govern the training and learning process of a machine learning model.

Area Under the ROC Curve (AUC)

A classification performance metric that evaluates a model's capability to distinguish between classes across multiple decision thresholds.

Chain-of-Thought

A prompt engineering and reasoning technique where models are encouraged to generate intermediate reasoning steps before arriving at a final answer.

Zero-Shot Learning

A technique enabling a model to perform tasks it was not explicitly trained on by leveraging knowledge acquired from related tasks.

Few-Shot Prompt Engineering

A technique where a model is supplied with a few demonstration examples to help it generalize and produce accurate predictions without retraining.

Instruction Tuning

A machine learning process where models are fine-tuned using explicit prompt-response instructions to follow human commands more effectively.

Batch Transform

A feature in Amazon SageMaker designed to generate inference predictions for large datasets in bulk when real-time response speeds are unnecessary.

Real-Time Inference

A SageMaker deployment option delivering instant, low-latency predictions for interactive applications requiring immediate responses.

Domain Adaptation Fine-Tuning

A customization method where a pre-trained foundation model is fine-tuned using domain-specific data to adapt it to specialized tasks or industries.

Mean Absolute Percentage Error (MAPE)

The mean of the absolute differences between actual and predicted values, divided by the actual values.

Recall

A performance metric measuring how effectively a classification model identifies all actual positive instances ($\text{True Positives}$) in a dataset.

Mean Squared Error (MSE)

A regression metric measuring the average squared difference between the predicted values and actual values.

Perplexity

An evaluation metric for language models that quantifies how well a probability model predicts a given sequence of words.

Amazon SageMaker Ground Truth

A data labeling service that incorporates human feedback and oversight to build high-quality, accurately labeled training datasets.

Amazon SageMaker Autopilot

An automated machine learning (AutoML) tool that automates model creation, training, and tuning with minimal manual effort.

Amazon SageMaker Pipelines

A dedicated workflow orchestration tool used to create consistent, repeatable, and scalable machine learning operations (MLOps).

Amazon Personalize

A managed machine learning service enabling developers to generate real-time personalized recommendations, custom re-ranking, and targeted marketing promotions.

AWS AI Service Cards

Documentation resources that describe AWS-developed foundation models and AI services, providing transparency details and benchmark performance data.

Low Bias

A condition indicating that a machine learning model does not make erroneous assumptions about the underlying relationships in the training data.

High Variance

A condition indicating that a model is overly sensitive to noise in the training data, leading to overfitting.

Transfer Learning

A machine learning technique where a model pre-trained on a broad base task is repurposed and fine-tuned for a related target task.

Amazon QuickSight

A cloud-powered business intelligence (BI) service used to build interactive dashboards, visual reports, and data analyses primarily on structured data.

Feature Selection

The process of selecting relevant data attributes or variables to train a predictive model effectively.

Model Validation

The evaluation phase where a model's performance and accuracy are tested to confirm suitability for production use cases before deployment.