L1 Flashcards
(104 cards)
1
Q
Overfitting - fix
A
Hyper parameter tuning
2
Q
Get started with LLMs and GenAI
A
Amazon Bedrock
Amazon SageMaker JumpStart
3
Q
Prevent confidential information in GenAI responses
A
Amazon Bedrock Guardrails
4
Q
maximum amount of text in LLM input
A
context window
5
Q
monitor and track the performance and usage of ML models
A
Amazon SageMaker Model Dashboard
6
Q
AWS service to detect potential bias via explain ability
A
Amazon SageMaker Clarify
7
Q
evaluate, compare, and select Foundation Models (FMs) quickly
A
Amazon SageMaker JumpStart
8
Q
human-in-the-loop to create training data
A
Amazon SageMaker Ground Truth
9
Q
image and video analysis for facial and object recognition
A
Amazon Rekognition
10
Q
Service to build, train, and deploy ML models, and customize for your needs
A
Amazon SageMaker
11
Q
extracts text from handwriting and scanned documents
A
Amazon Textract
12
Q
service to build and deploy AI applications with access to foundation LLM models.
A
Amazon Bedrock
13
Q
NLP service to uncover intent, sentiment and relationships
A
Amazon Comprehend
14
Q
time-series forecasts
A
Amazon Forecast
15
Q
pay for actual usage, without an upfront payment or long-term contract.
A
on-demand pricing
16
Q
reserves a predictable capacity in advance for a discounted rate
A
provisioned throughput
17
Q
spare EC2 capacity at reduced rates
A
spot instances
18
Q
lower rates for a long-term commitment
A
reserved instances
19
Q
prompt to avoid certain outputs or behaviors
A
Negative prompting
20
Q
prompting to generate content without having seen any examples, relying solely on its general understanding
A
Zero-shot Prompting
21
Q
prompt providing a few examples of a task
A
Few-shot Prompting
22
Q
prompt a complex question as a series of intermediate steps
A
Chain-of-thought prompting
23
Q
Foundation Models - supervised or self-supervised learning
A
creation - self-supervised learning
fine-tuning - supervised learning
24
Q
The hyperparameter to control the creativity / randomness of LLM responses
A
Temperature - higher temperature = higher creativity
25
minimum Availability Zones (AZ) in a AWS Region
3
26
Minimum discrete data centers in Each Availability Zone (AZ)
1
27
Creating large language models from foundation model
Fine tuning
28
Adjusting the temperature to set creativity level of LLM
Hyper parameter tuning
29
What type of learning is classification
Supervised learning
30
What type of learning is regression, and time series
Supervised learning
31
What type of learning is neural networks and object detection
Supervised learning
32
What type of learning is clustering and dimensionality reduction
Unsupervised learning
33
What type of learning is association rule/relationship learning
unsupervised learning
34
What type of learning is anomaly, detection, and fraud detection
Unsupervised learning
35
machine learning where a model is trained on labeled data to map
supervised learning
36
machine learning where a model identifies patterns and structures in unlabeled data
unsupervised learning
37
training robots and chatbots where direct feedback is available, by providing rewards or penalties based on its actions
Reinforcement learning (vs supervised/unsupervised learning)
38
Building a domain expert model using a foundation model
Continued Pre-Training on domain specific data
Domain Adaptation Fine-Tuning
39
enterprise search service powered by ML
Amazon Kendra
40
Service to simplify data preparation (aggregate and prepare tabular) and feature engineering
Amazon SageMaker Data Wrangler
41
retrieve information from a knowledge base, then use LLM to generate a response
RAG
42
convert audio input into text
Amazon Transcribe
43
models and algorithms capable of creating new content including text, images, audio and video based on patterns learned from existing data
Generative AI
44
Service for no-code creation of ML models
Amazon SageMaker Canvas
45
repository to store, and manage features for ML models
Amazon SageMaker Feature Store
46
Service to Integrate RAG + LLM in chatbot for accuracy
Knowledge Bases for Amazon Bedrock
47
two competing neural networks training each other
Generative adversarial network (GAN)
48
having insights into model’s decision-making process
Explainability
49
models with high explain ability
Decision tree, linear and logistic regression
50
models with low explain ability
Neural networks, deep learning, and Transformers
51
Generate additional examples to balance the dataset
Augmentation
52
prevent overfitting by adding a penalty to the loss function
regularization
53
update model with new data over time
incremental training
54
models to learn from each other by sharing the latest data insights
transfer learning
55
comparing the outputs of models to select the best suited model for a use case
Model evaluation
56
a model generating an output (response) from a given input (prompt)
Inference
57
AI-based applications that can complete complex tasks
Agents
58
GenAI-based applications that can complete complex tasks
Agents for Amazon Bedrock
59
analyze logs, metrics, and documentation to provide insights
Amazon Q
60
Service for Image generation using natural language prompts
Amazon Titan
61
MLflow
open-source platform to manage machine learning lifecycle - track experiments, package models, and manage deployments
62
use a self-attention mechanism and implement contextual embeddings
Transformer models
63
cleaning and preprocessing the data for training
Data preparation aka Data pre-processing
64
examining the data through statistical summaries and visualizations to identify patterns, detect anomalies etc.
Exploratory Data Analysis (EDA)
65
7 stages of the ML lifecycle
1. Problem Definition
2. Data Collection
3. Exploratory Data Analysis (EDA)
4. Data Preprocessing
5. Model Selection & Training
6. Evaluation & Optimization
7. Deployment & Monitoring
66
tailored experiences based on user behavior
Amazon Personalize
67
synthesize natural-sounding human speech
Amazon Polly
68
GenAI Service for automated code suggestions
Amazon Q Developer
69
Amazon Q is developed on top of which AWS service
Amazon Bedrock
70
The hyperparameter to limit the LLMs next word choices by cumulative probability threshold
Top-P (P=probability)
71
The hyperparameter to limit the LLMs next word choices to the most probable tokens
Top-K
72
similarity searches and fast index lookups
vector db
73
AWS managed vector DBs
Amazon OpenSearch Service
74
Responsible for security of the cloud - AWS or customer
AWS
75
Responsible for security in the cloud - AWS or customer
Customer
76
BLEU (Bilingual Evaluation Understudy) score
Evaluating the quality of translation from one language to another
77
Asynchronous inference
smaller payloads without requiring real-time responses
78
Batch inference
larger payloads without requiring real-time responses
79
AWS service to provide high-quality translation on demand
Amazon Translate
80
service that provides guidance for following AWS best practices
AWS Trusted Advisor
81
service that logs all API calls for auditing
AWS CloudTrail
82
service to access metrics, logs, and alarms of AWS resources
AWS CloudWatch
83
prompt engineering
Effective Instructions, Context, Input data, Output Indicator
84
Deep Learning
ML based on neural networks used for Image recognition and Natural language processing (NLP)
85
Service that monitors the quality of Amazon SageMaker ML models
Amazon SageMaker Model Monitor
86
Metric for performance of classification models
Confusion matrix
87
Metric for performance of regression models
Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE)
88
metric for relation between different variables or features in a dataset
Correlation matrix
89
True or false - AWS cost is directly proportional to the number of tokens in the input
True
90
converting data—such as text, images, etc.—into numerical representations (vectors) that capture semantic meaning
Embedding
91
bias occurs when the data used to train the model does not accurately reflect the diversity of the real-world population
Sampling bias
92
involves inaccuracies in data collection, such as faulty equipment
Measurement bias
93
how to audit requests and responses during model invocations in Amazon Bedrock
enable model invocation logging within Amazon Bedrock
94
training iterations over the dataset is called
Epoch
95
True or false - Increasing the number of epochs of ML model can improve accuracy
Yes, up to a certain extent
96
can on-demand compute be provisioned to test and deploy fine-tuned custom models
Provisioned Throughput mode is mandatory and AWS does not allow to use options like on demand
97
console environment to experiment with running inference on different models before deciding
Amazon Bedrock playground
98
algorithm specifically designed to capture the contextual meaning of words by looking at both the words that come before and after them (bidirectional context)
Bidirectional Encoder Representations from Transformers (BERT)
99
what explain ability method ensures that everyone (or every feature) gets the right credit for their influence on the final outcome, for a single output of the model.
Shapley Values (Game-Theoretic Feature Attribution)
100
what explain ability method shows how adjusting a single feature (sugar in the cookies, or a feature in a model) impacts the final prediction across all outputs of the model
Partial Dependence Plots (PDP) (Feature Impact Visualization)
101
characters that the model processes as discrete units of text
tokens
102
service for integrating human review into the decision-making process of models post-prediction
Amazon Augmented AI (A2I)
103
Image generation models type for Dall-E, Firefly
Diffusion model
104
