AI P Flashcards
1
Q
Supervised Learning
A
Uses labeled data to train models for tasks like classification and spam detection.
2
Q
Unsupervised Learning
A
Uses unlabeled data to detect patterns, useful for customer segmentation and anomaly detection.
3
Q
Self-Supervised Learning
A
Generates its own labels to predict and infer missing information, useful for NLP tasks.
4
Q
Semi-Supervised Learning
A
Uses a mix of labeled and unlabeled data to learn and cluster data.
5
Q
Reinforcement Learning
A
Learns from interacting with the environment and receiving feedback, ideal for recommendation systems and self-driving cars.
6
Q
Underfitting
A
Occurs when a model doesn’t learn enough from the training data, leading to poor performance.
7
Q
Overfitting
A
Occurs when a model learns too many details, including noise and outliers, leading to poor performance on new data.
8
Q
Foundation Models
A
Super large models trained on vast amounts of data, adaptable for various tasks.
9
Q
Fine-Tuning
A
Customizing a pre-trained model for a specific task or dataset by training it further.
10
Q
Exploratory Data Analysis (EDA)
A
Examining and understanding a dataset before complex analysis or modeling.
11
Q
Feature Engineering
A
Transforming raw data into meaningful features to enhance model predictions.
12
Q
Hyperparameters
A
Settings selected before training a model, such as learning rate, batch size, and number of epochs.
13
Q
Parameters
A
Values automatically learned by the model during training, such as weights and biases in neural networks.
14
Q
Classification Metrics
A
Metrics used to evaluate classification models, including accuracy, precision, recall, and F1 score.
15
Q
Regression Metrics
A
Metrics used to evaluate regression models, including mean absolute error, root mean squared error, and R squared.
16
Q
Amazon Rekognition
A
Allows machines to interpret images and videos using machine learning. Can be customized for content moderation and specific object detection.
17
Q
Amazon Textract
A
“Automatically extracts text and data from scanned documents for digitization and analysis.
18
Q
Amazon Comprehend
A
Understands and analyzes the meaning and sentiment behind text.
19
Q
Amazon Translate
A
Automatically translates text between multiple languages.
20
Q
Amazon Polly
A
Converts text into natural sounding speech.
21
Q
Amazon Transcribe
A
Converts spoken language into written text, with options for custom vocabularies and transcripts.
22
Q
Amazon Lex
A
Builds conversational interfaces using NLP and automatic speech recognition.
23
Q
Amazon Forecast
A
Predicts future trends by identifying historical patterns.
24
Q
Amazon Kendra
A
Builds a search engine by crawling documents and understanding context for relevant results.
25
Amazon Personalize
Provides tailored experiences by analyzing user behaviors, preferences, and trends.
26
Amazon SageMaker
A fully managed service that helps you build, train, and deploy machine learning models quickly.
27
SageMaker Studio
An IDE that simplifies all your machine learning needs.
28
Data Wrangler
A tool that simplifies the process of preparing and cleaning data.
29
Feature Store
A centralized repository for storing and managing machine learning features to avoid redundancy.
30
SageMaker Pipelines
Automates and orchestrates ML workflows by breaking the process into simple steps.
31
AutoML
Automates the process of model selection and hyperparameter tuning.
32
Real-Time Endpoints
Provides synchronous inferencing with low latency, ideal for time-sensitive applications.
33
Asynchronous Inference
Offers moderate to high latency, suited for complex models like image and video analysis.
34
Batch Transform
Also known as batch inference, it has high latency and is suited for processing large datasets.
35
Serverless Deployment
Offers low to moderate latency, suitable for scalable applications with variable workloads.
36
Tokens
Fundamental building blocks of language models representing individual words or parts of words.
37
Context Window
The limit on how much text the model can analyze at once defining the maximum number of tokens the model can remember.
38
Chunking
Dividing large pieces of text into smaller manageable parts for efficient processing.
39
Embeddings
Represent data in a way that machines can understand by converting words or images into numbers reflecting their meaning and relationships.
40
Vectors
Numerical arrays of embeddings representing points or positions in space.
41
Transformer-Based LLMs
Process input data using neural networks to generate human-understandable output.
42
Foundation Models
Provide a base for further specialization in generative AI.
43
Multimodal Models
Work across text image and audio data.
44
Diffusion Models
Generate realistic images by refining noisy inputs.
45
Prompt Engineering
Designing inputs or prompts to guide generative AI models to produce a desired output.
46
Data Selection
Choosing relevant and high-quality data to train the model.
47
Model Selection
Choosing a suitable model architecture based on problem requirements.
48
Pre-Training
Training the model on large-scale general datasets to learn basic patterns and structures.
49
Epoch
A complete iteration through a training dataset during model training.
50
Fine-Tuning
Customizing the pre-trained model with task-specific data.
51
Evaluation Deployment and Feedback
The final stages of the foundation model lifecycle.
52
Generative AI Models
Create realistic images and videos from text descriptions can be used as chatbots for translation product recommendations code generation and text summarization.
53
Hallucinations
When the model generates content that may sound convincing but is factually incorrect.
54
Non-Determinism
When the same input can yield different outputs each time.
55
Bias
Models trained on biased data can produce biased or unfair content.
56
Diffusion Models
Generate realistic images by refining noisy inputs.
57
Multimodal Models
Work across text image and audio data.
58
Amazon SageMaker
A fully managed service that helps you build train and deploy machine learning models quickly.
59
Amazon Bedrock
A fully managed service that allows you to build and scale AI applications using foundation models.
60
PartyRock
An interactive environment where developers can experiment with and deploy generative AI models.
61
Amazon Q
An innovative AI service that helps users generate visual insights from complex business data.
62
Jumpstart
Provides access to built-in algorithms pre-trained models and machine learning solutions.
63
Real-Time Inference
Provides fast response times for time-sensitive applications.
64
Asynchronous Inference
Offers moderate to high latency suited for complex models.
65
Batch Transform
Processes large datasets with high latency.
66
Serverless Deployment
Offers scalable applications with variable workloads.
67
Responsiveness
How quickly a model reacts to input typically requiring more compute power for faster responses.
68
Availability
Ensuring a service is up and running at all times requiring duplicate resources and failover systems.
69
Redundancy
Having backup systems to prevent downtime improving reliability but increasing infrastructure costs.
70
Token-Based Pricing
Paying for the number of tokens processed by the model offering flexibility but variable usage.
71
Provisioned Throughput
Reserved capacity for steady performance requiring pre-allocated resources.
72
Custom Models
Provide flexibility and precision for business needs requiring significant compute power and storage.
73
Cost
The expense associated with using a foundation model including computational and storage costs.
74
Modality
The type of data the model processes such as text image or audio.
75
Latency
How quickly a model processes requests.
76
Model Size and Complexity
The scale and intricacy of the model affecting performance and cost.
77
Multilingual Capabilities
The ability of a model to understand and generate multiple languages.
78
Customization Options
Ways to tailor a model to specific needs such as pre-training fine-tuning RAG and in-context learning.
79
Inference Parameters
Settings that control how a trained model generates outputs during the inference phase.
80
Temperature
Adjusts the creativity of the model's outputs; higher values mean more diverse outputs lower values mean more deterministic responses.
81
Top P
The percentage of most likely candidates considered for the next token.
82
Top K
The number of most likely candidates considered for the next token.
83
Input and Output Length
The length of input or output text affecting costs and computational demands.
84
RAG (Retrieval-Augmented Generation)
Combines a foundation model with an external knowledge source to improve response accuracy.
85
Knowledge Base
A structured repository of information that a foundation model can reference during inference.
86
Vector Databases
Use vector search algorithms to index and query vector embeddings based on their similarity.
87
Agents for Amazon Bedrock
Automate multi-step workflows integrating with RAG for real-time context-aware decision-making.
88
Context
Frames the task by providing relevant background information.
89
Instruction
Directs the model and sets expectations.
90
Negative Prompts
Tell the model what not to include in an output.
91
Model Latent Space
The knowledge and patterns models use to generate responses.
92
Zero-Shot Prompting
Providing the model with a task without offering any examples.
93
Single-Shot Prompting
Providing the model with one example.
94
Few-Shot Prompting
Providing the model with multiple examples.
95
Chain-of-Thought Prompting
Asking the model to break down a complex problem into smaller logical steps.
96
Prompt Templates
Designing a template to translate user input into instructions for a language model.
97
Pre-Training
The initial stage where a model learns from vast amounts of unstructured data to develop general capabilities.
98
Fine-Tuning
Customizing a pre-trained model with task-specific data.
99
Continuous Pre-Training
Further training a language model to learn new information while retaining what it has already learned.
100
Instruction Tuning
Further training a foundation model to excel at following explicit instructions.
101
Adapting Models for Specific Domains
Further training a general-purpose AI model on data specific to a particular industry or field.
102
Transfer Learning
Taking a model pre-trained on one task and fine-tuning it for a new related task.
103
Data Curation
Selecting and organizing relevant accurate and high-quality data.
104
Data Governance
Policies processes and practices to ensure the quality accuracy and ethical use of data.
105
Data Size
The total amount of data used to train the model.
106
Data Representativeness
How well the training data reflects the diversity and characteristics of the real-world population.
107
Data Labeling
Annotating data with tags or categories to help the model understand specific patterns.
108
Reinforcement Learning from Human Feedback (RLHF)
Training a reward model with direct human feedback to produce better responses.
109
Human Evaluation
Involves people assessing a model's outputs based on specific criteria.
110
Benchmark Datasets
Pre-built collections of labeled data used to test model performance against industry standards.
111
ROUGE (Recall-Oriented Understudy for Gisting Evaluation)
Measures overlap between generated and reference texts commonly used for summarization tasks.
112
BLEU (Bi-Lingual Evaluation Understudy)
Evaluates how closely a generated translation matches a reference by comparing word sequences.
113
BERTScore
Uses pre-trained BERT models to compare semantic similarity between generated and reference texts.
114
Productivity
Refers to how efficiently the model performs tasks and generates high-quality outputs with minimal human intervention.
115
User Engagement
Looks at how often and how deeply users interact with the model.
116
Task Engineering
Refers to how effectively the model can complete specific tasks that align with business objectives.
117
Responsible AI
Ensuring that ML algorithms are ethical transparent and trustworthy.
118
Transparency
Being able to see how a machine learning model makes decisions including access to training data algorithms and processes.
119
Explainability
Understanding why a model is making a specific prediction by breaking down the reasoning behind its decision.
120
Fairness
Ensuring that AI systems are unbiased and equitable.
121
Controllability
The ability to control and adjust AI systems as needed.
122
Veracity and Robustness
Ensuring the accuracy and reliability of AI systems.
123
Safety
Ensuring that AI systems do not cause harm.
124
Privacy and Security
Protecting sensitive information and ensuring data security.
125
Governance
Implementing policies and practices to oversee AI systems.
126
Bias
Systematic errors in predictions leading to underfitting or overfitting.
127
High Bias
Consistently missing the target in the same direction leading to underfitting.
128
High Variance
Predictions that vary widely leading to overfitting.
129
Measurement Bias
Bias arising from inaccurate measurements.
130
Sampling Bias
Bias arising from non-representative samples.
131
Confirmation Bias
Bias arising from favoring information that confirms pre-existing beliefs.
132
Observer Bias
Bias arising from the observer's expectations influencing the outcome.
133
Hallucinations
Generative AI presenting fabricated details as plausible facts.
134
Exposure
AI models unintentionally revealing sensitive information.
135
Prompt Leaking
A model accidentally exposing details about the specific prompt it received.
136
Human-Centered Design (HCD)
Designing AI systems with real human needs in mind focusing on safety and fairness.
137
Amazon SageMaker Clarify
Helps detect and mitigate bias in datasets and models providing insights into how predictions are made.
138
Amazon SageMaker GroundTruth
A tool for labeling training data through human annotators or automated workflows ensuring accurate and well-labeled datasets.
139
Model Cards
Documents information about machine learning models including their purpose performance metrics limitations and ethical considerations.
140
Model Monitor
Continuously tracks model performance in real-world environments detecting issues like data drift and bias.
141
Amazon Augmented AI (A2I)
Enables easy setup of workflows for human review and correction of AI outputs acting as a safety net for AI.
142
Amazon Bedrock Guardrails
Filters and refines generated content to prevent harmful or inappropriate elements ensuring safe and ethical AI usage.
143
HIPAA (Health Insurance Portability and Accountability Act)
A US regulation designed to protect the protected health information of US citizens and their medical information.
144
AWS Identity and Access Management (IAM)
Allows you to securely manage user access and permissions for your AWS resources and services.
145
AWS Key Management Service (KMS)
Performs cryptographic operations to support the confidentiality of data stored and transmitted on AWS.
146
Amazon Elastic Load Balancer (ELB)
Supports the availability of applications by balancing traffic and mitigating denial-of-service attacks.
147
AWS Systems Manager
Provides control and visibility into AWS resources but does not provide access control at the level of IAM.
148
AWS Control Tower Guardrails
Helps mitigate the risk of prompt injection by enforcing input validation and ensuring only approved prompt templates are used.
149
Data Privacy Regulations
Laws and regulations designed to protect personal data such as GDPR and PCI-DSS.
150
Data Sovereignty Requirements
Laws and regulations that dictate how data is stored and processed within a specific country.
151
AWS CloudTrail
Logs all API activity across AWS services including successful and failed API calls for auditing and compliance purposes.
152
ISO 27001
Specifies the requirements for establishing implementing maintaining and continually improving an information security management system (ISMS).
153
ISO 27002
Provides guidelines and best practices for implementing the requirements of ISO 27001 including security controls.
