Choosing an AWS machine learning service
Pick the right ML services and frameworks to support your work
|
Purpose
|
Help determine which AWS ML services are the best fit for your needs.
|
|
Last updated
|
May 3, 2024
|
|
Covered services
|
|
Introduction
At its most basic, machine learning (ML) is designed to provide
digital tools and services to learn from data, identify patterns,
make predictions, and then act on those predictions. Almost all
artificial intelligence (AI) systems today are created using ML. ML
uses large amounts of data to create and validate decision logic.
This decision logic forms the basis of the
AI model.
Scenarios where AWS machine learning services may be applied
include:
-
Specific use cases — AWS machine learning services
can support your AI powered use cases with a broad range of pre-built algorithms, models,
and solutions for common use cases and industries. You have a choice of 23 pre-trained
services, including Amazon Personalize, Amazon Kendra, and Amazon Monitron.
-
Customizing and scaling machine learning — Amazon SageMaker
is designed to help you build, train, and deploy ML models for any use case. You can build
your own or access open source foundational models on AWS through Amazon SageMaker and Amazon Bedrock.
-
Accessing specialized infrastructure — Use the ML
frameworks and infrastructure provided by AWS when you require even greater flexibility
and control over your machine learning workflows, and are willing to manage the underlying
infrastructure and resources yourself.
This decision guide will help you ask the right questions, evaluate
your criteria and business problem, and determine which services are
the best fit for your needs.
Understand
As organizations continue to adopt AI and ML technologies, the
importance of understanding and choosing among AWS ML services is an
on-going challenge.
AWS provides a range of ML services designed to help organizations
to build, train, and deploy ML models more quickly and easily. These
services can be used to solve a wide range of business problems such
as customer churn prediction, fraud detection, and image and speech
recognition.
Before diving deeper into AWS ML services, let's look at the
relationship between AI and ML.
-
At a high level, artificial intelligence is
a way to describe any system that can replicate tasks that
previously required human intelligence. Most AI use cases are
looking for a probabilistic outcome—making a prediction or
decision with a high degree of certainty, similar to human
judgement.
-
Almost all AI systems today are created using machine
learning. ML uses large amounts of data to create and
validate decision logic, which is known as a model.
-
Classification AI is a subset of ML that recognizes patterns to
identify something. Predictive AI is a subset of ML that
predicts future trends based on statistical patterns an
historical data.
-
Finally, generative AI is a subset of deep learning that can
create new content and ideas, like conversations, stories,
images, videos, and music. Generative AI is powered by very
large models that are pretrained on vast corpora of data, called
the Foundation Models or
FMs. Amazon Bedrock
is a fully managed service that offers
a choice of high-performing FMs for building and scaling
generative AI
applications. Amazon Q Developer and
Amazon Q Business are generative-AI powered
assistants for specific use cases.
This guide is designed primarily to cover services in
the Classification AI and Predictive
AI machine learning categories.
In addition, AWS offers specialized, accelerated hardware for high
performance ML training and inference.
-
Amazon EC2 P5 instances are equipped with NVIDIA H100
Tensor Core GPUs, which are well-suited for both training and
inference tasks in machine
learning. Amazon EC2 G5 instances feature up to 8 NVIDIA A10G
Tensor Core GPUs, and second generation AMD EPYC processors, for
a wide range of graphics-intensive and machine learning use
cases.
-
AWS Trainium is the second-generation ML
accelerator that AWS has purpose-built for deep learning (DL)
training of 100B+ parameter models.
-
AWS Inferentia2-based Amazon EC2 Inf2 instances are designed to deliver high
performance at the lowest cost in Amazon EC2 for your DL and
generative AI inference applications.
Consider
When solving a business problem with AWS ML services, consideration
of several key criteria can help ensure success. The following
section outlines some of the key criteria to consider when choosing
a ML service.
- Problem definition
-
Problem definition
The first step in the ML lifecycle is to frame the business problem.
Understanding the problem you are trying to solve is essential for
choosing the right AWS ML service, as different services are
designed to address different problems. It is also important to
determine whether ML is the best fit for your business problem.
Once you have determined that ML is the best fit, you can start by
choosing from a range of purpose-built AWS AI services (in areas
such as speech, vision and documents).
Amazon SageMaker provides fully managed infrastructure if you need
to build and train your own models. AWS offers an array of advanced
ML frameworks and infrastructure choices for the cases where you
require highly customized and specialized ML models. AWS also offers
a broad set of popular foundation models for building new
applications with generative AI.
- ML algorithm
-
ML algorithm
Choosing the ML algorithm for the business problem you are trying to
solve depends on the type of data you are working with, as well as
the desired outcomes. The following information outlines how each of
the major AWS AI/ML service categories empowers you to work with its
algorithms:
-
Specialized AI services: These services offer a limited ability
to customize the ML algorithm, as they are pre-trained models
optimized for specific tasks. You can typically customize the
input data and some parameters, but do not have access to the
underlying ML models or the ability to build your own models.
-
Amazon SageMaker: This service provides the most flexibility and
control over the ML algorithm. You can use SageMaker to build
custom models using your own algorithms and frameworks, or use
pre-built models and algorithms provided by AWS. This allows for
a high degree of customization and control over the ML process.
-
Lower-level ML frameworks and infrastructure: These services
offer the most flexibility and control over the ML algorithm.
You can use these services to build highly customized ML models
using their own algorithms and frameworks. However, using these
services requires significant ML expertise and may not be
feasible for all every use case.
- Security
-
Security
If you need a private endpoint in your VPC, your options will vary
based on the layer of AWS ML services you are using. These include:
-
Specialized AI services: Most specialized AI services do not
currently support private endpoints in VPCs. However, Amazon Rekognition Custom Labels and Amazon Comprehend
Custom can be accessed using VPC endpoints.
-
Core AI services: Amazon Translate, Amazon Transcribe, and
Amazon Comprehend all support VPC endpoints.
-
Amazon SageMaker: SageMaker provides built-in support for VPC
endpoints, allowing you to deploy their trained models as an
endpoint accessible only from within their VPC.
-
Lower-level ML frameworks and infrastructure: You can deploy
your models on Amazon EC2 instances or in containers within your
VPC, providing complete control over the networking
configuration.
- Latency
-
Latency
Higher-level AI services, such as Amazon Rekognition and Amazon Transcribe, are designed to
handle a wide variety of use cases and
offer high performance in terms of speed. However, they might not
meet certain latency requirements.
If you are using lower-level ML frameworks and infrastructure, we
recommended leveraging Amazon SageMaker. This option is generally
faster than building custom models due to its fully managed service
and optimized deployment options. While a highly optimized custom
model may outperform SageMaker, it will require significant
expertise and resources to build.
- Accuracy
-
Accuracy
The accuracy of AWS ML services varies based on the specific use
case and level of customization required. Higher-level AI services,
such as Amazon Rekognition, are built on pre-trained models that
have been optimized for specific tasks and offer high accuracy in
many use cases.
In some cases, you can choose to use Amazon SageMaker, which
provides a more flexible and customizable platform for building and
training custom ML models. By building your own models, you may be
able to achieve even higher accuracy than what is possible with
pre-trained models.
You can also choose to use ML frameworks and infrastructure, such as
TensorFlow and Apache MXNet, to build highly customized models that
offer the highest possible accuracy for your specific use case.
- AWS and responsible AI
-
AWS and responsible AI
AWS builds foundation models (FMs) with responsible AI in mind at
each stage of its development process. Throughout design,
development, deployment, and operations we consider a range of
factors including:
-
Accuracy (how closely a summary matches the underlying document;
whether a biography is factually correct)
-
Fairness, (whether outputs treat demographic groups similarly)
-
Intellectual property and copyright considerations
-
Appropriate usage (filtering out user requests for legal advice,
or medical diagnoses, or illegal activities)
-
Toxicity (hate speech, profanity, and insults)
-
Privacy (protecting personal information and customer prompts)
AWS builds solutions to address these issues into the processes used
for acquiring training data, into the FMs themselves, and into the
technology used to pre-process user prompts and post-process
outputs.
Choose
Now that you know the criteria by which you will be evaluating your ML service options, you are ready to choose which AWS ML service is
right for your organizational needs.
The following table highlights which ML services are optimized for which circumstances. Use it to help determine the AWS ML service that is the
best fit for your use case.
|
Categories
|
When would you use it?
|
What is it optimized for?
|
Related AI/ML services or
environments
|
|
Specific use cases
These artificial intelligence services are intended to
meet specific needs. They include personalization,
forecasting, anomaly detection, speech transcription, and
others. Since they are delivered as services, they can be
embedded into applications without requiring any ML
expertise.
|
Use the AI services provided by AWS when you require
specific, pre-built functionalities to be integrated into
your applications, without the need for extensive
customizations or machine learning expertise. These services
are designed to be easy to use and do not require much
coding or configuration.
|
These services are designed to be easy to use and do not
require much coding, configuration, or ML expertise.
|
Amazon Augmented AI
Amazon CodeGuru
Amazon Comprehend
Amazon Comprehend Medical
Amazon DevOps Guru
Amazon Forecast
Amazon Kendra
Amazon Lex
Amazon Personalize
Amazon Polly
Amazon Rekognition
Amazon Textract
Amazon Transcribe
Amazon Translate
|
|
ML services
These services can be used to develop customized machine
learning models or workflows that go beyond the pre-built
functionalities offered by the core AI services.
|
Use these services when when you need more customized
machine learning models or workflows that go beyond the
pre-built functionalities offered by the core AI services.
|
These services are optimized for building and training
custom machine learning models, large-scale training on
multiple instances or GPU clusters, more control over
machine learning model deployment, real-time inference, and
for building end-to-end workflows.
|
Amazon SageMaker
Amazon SageMaker JumpStart
SageMaker Studio
SageMaker Canvas
SageMaker Studio Lab
SageMaker Ground Truth
PyTorch on AWS
Apache MxNet
Hugging Face
TensorFlow on AWS
|
|
Infrastructure
To deploy machine learning in production, you need
cost-effective infrastructure, which Amazon enables with
AWS-built silicon.
|
Use when you want to achieve the lowest cost for training
models and need to run inference in the cloud.
|
Optimized for supporting the cost-effective deployment of
machine learning.
|
AWS Trainium
AWS Inferentia and Inferentia2
Amazon SageMaker HyperPod
|
|
Tools and associated
services
These tools and associated services are designed to help
you ease deployment of machine learning.
|
These services and tools are designed to help you accelerate
deep learning in the cloud, providing Amazon machine images,
docker images and entity resolution.
|
Optimized for helping you accelerate deep learning in the
cloud.
|
AWS Deep Learning AMIs
AWS Deep Learning Containers
AWS Entity Resolution
|
Use
Now that you have a clear understanding of the criteria you need to
apply in choosing an AWS ML service, you can select which AWS AI/ML
service(s) are optimized for your business needs.
To explore how to use and learn more about the service(s) you have
chosen, we have provided three sets of pathways to explore how each
service works. The first set of pathways provides in-depth
documentation, hands-on tutorials, and resources to get started with
Amazon Comprehend, Amazon Textract, Amazon Translate, Amazon Lex,
Amazon Polly, Amazon Rekognition, and Amazon Transcribe.
- Amazon Comprehend
-
-
Get started with Amazon Comprehend
Use the Amazon Comprehend console to create and run an asynchronous entity detection job.
Get started
with the tutorial »
-
Analyze insights in text with Amazon Comprehend
Learn how to use Amazon Comprehend to analyze and derive insights from text.
Get
started with the tutorial »
-
Amazon Comprehend Pricing
Explore information on Amazon Comprehend pricing and examples.
Explore the guide
»
- Amazon Textract
-
-
Getting Started with Amazon Textract
Learn how Amazon Textract can be used with formatted text to detect words and lines of words that are located close to each other, as well as analyze a document for items such as related text, tables, key-value pairs, and selection elements.
Explore the guide »
-
Extract text and structured data with Amazon Textract
Learn how to use Amazon Textract to extract text and structured data from a document.
Get started with the tutorial »
-
AWS Power Hour: Machine Learning
Dive into Amazon Textract in this episode, spend time in the AWS Management Console,
and review code samples that will help you understand how to make the
most of service APIs.
Watch the video
»
- Amazon Translate
-
-
Getting started with Amazon Translate using the console
The easiest way to get started with Amazon Translate is to use the console to translate some text. Learn how to translate up to 10,000 characters using the console.
Explore the guide »
-
Translate Text Between Languages in the Cloud
In this tutorial example, as part of an international luggage manufacturing firm, you need to understand what customers are saying about your product in reviews in the local market language - French.
Get started with the tutorial »
-
Amazon Translate pricing
Explore Amazon Translate pricing, including Free Tier - which provides 2 million characters per month for 12 months.
Explore the guide »
- Amazon Lex
-
-
Amazon Lex V2 Developer Guide
Explore information about getting started, how it works, and pricing information for Amazon Lex V2.
Explore the guide »
-
Introduction to Amazon Lex
We introduce you to the Amazon Lex conversational service, and walk you through examples that show you how to create a bot and deploy it to different chat services.
Take the course » (sign-in required)
-
Exploring Generative AI in conversational experiences
Explore the use of generative AI in conversation experiences.
Read the blog »
- Amazon Polly
-
-
What is Amazon Polly?
Explore a complete overview of the cloud service that converts text into lifelike speech, and can be used to develop applications to increase your customer engagement and accessibility.
Explore the guide »
-
Highlight text as it’s being spoken using Amazon Polly
We introduce you to approaches for highlighting text as it’s being spoken to add visual capabilities to audio in books, websites, blogs, and other digital experiences.
Read the blog »
-
Create audio for content in multiple languages with the same TTS voice persona in Amazon Polly
We explain Neural Text-to-Speech (NTTS) and discuss how a broad portfolio of available voices, providing a range of distinct speakers in supported languages, can work for you.
Read the blog »
- Amazon Rekognition
-
-
What is Amazon Rekognition?
Explore how you can use this service to add image and video analysis to your applications.
Explore the guide »
-
Hands-on Rekognition: Automated Image and Video Analysis
Learn how facial recognition works with streaming video, along with code examples and key points at a self-guided pace.
Get started with the tutorial »
-
Amazon Rekognition FAQs
Learn the basics of Amazon Rekognition and how it can help you improve your deep learning and visually analyze your applications.
Read the FAQs »
- Amazon Transcribe
-
-
What is Amazon Transcribe?
Explore the AWS automatic speech recognition service using ML to convert audio to text. Learn how to use this service as a standalone transcription or add speech-to-text capability to any application.
Explore the guide »
-
Amazon Transcribe Pricing
We introduce you to the AWS pay-as-you-go transcription, including
custom language model options and the Amazon Transcribe Free Tier.
Explore the guide
»
-
Create an audio transcript with Amazon Transcribe
Learn how to use Amazon Transcribe to create a text transcript of recorded audio files using a real-world use case scenario for testing against your needs.
Get started with the tutorial »
-
Build an Amazon Transcribe streaming app
Learn how to build an app to record, transcribe, and translate live audio in real-time, with results emailed directly to you.
Explore the guide »
The second set of AI/ML AWS service pathways provide in-depth
documentation, hands-on tutorials, and resources to get started with
the services in the Amazon SageMaker family.
- SageMaker
-
-
How Amazon SageMaker works
Explore the overview of machine learning and how SageMaker works.
Explore the guide »
-
Getting started with Amazon SageMaker
Learn how to join an Amazon SageMaker Domain, giving you access to Amazon SageMaker Studio and RStudio on SageMaker.
Explore the guide »
-
Use Apache Spark with Amazon SageMaker
Learn how to use Apache Spark for preprocessing data and SageMaker for model training and hosting.
Explore the guide »
-
Use Docker containers to build models
Explore how Amazon SageMaker makes extensive use of Docker containers for build and runtime tasks. Learn how to deploy the pre-built Docker images for its built-in algorithms and the supported deep learning frameworks used for training and inference.
Explore the guide »
-
Machine learning frameworks and languages
Learn how to get started with SageMaker using the Amazon SageMaker Python SDK.
Explore the guide »
- SageMaker Autopilot
-
-
Create an Amazon SageMaker Autopilot experiment for tabular data
Learn you how to create an Amazon SageMaker Autopilot experiment to explore, pre-process, and train various model candidates on a tabular dataset.
Explore the guide »
-
Automatically create machine learning models
Learn how to use Amazon SageMaker Autopilot to automatically build, train, and tune a ML model, and deploy the model to make predictions.
Get started with the tutorial »
-
Explore modeling with Amazon SageMaker Autopilot with these example notebooks
Explore example notebooks for direct marketing, customer churn prediction and how to bring your own data processing code to Amazon SageMaker Autopilot.
Explore the guide »
- SageMaker Canvas
-
-
Get started using Amazon SageMaker Canvas
Learn how to get started with using SageMaker Canvas.
Explore the guide »
-
Generate machine learning predictions without writing code
This tutorial explains how to use Amazon SageMaker Canvas to build ML models and generate accurate predictions without writing a single line of code.
Get started with the tutorial »
-
Dive deeper into SageMaker Canvas
Explore an in-depth look at SageMaker Canvas and its visual, no code ML capabilities.
Read the blog »
-
Use Amazon SageMaker Canvas to make your first ML Model
Learn how to use Amazon SageMaker Canvas to create an ML model to assess customer retention, based on an email campaign for new products and services.
Get started with the lab »
- SageMaker Data Wrangler
-
-
Getting started with Amazon SageMaker Data Wrangler
Explore how set up SageMaker Data Wrangler and then provides a walkthrough using an existing example dataset.
Explore the guide »
-
Prepare training data for machine learning with minimal code
Learn how to prepare data for ML using Amazon SageMaker Data Wrangler.
Get started with the tutorial »
-
SageMaker Data Wrangler deep dive workshop
Learn how to apply appropriate analysis types on your dataset to detect anomalies and issues, use the derived results/insights to formulate remedial actions in the course of transformations on your dataset, and test the right choice and sequence of transformations using quick modeling options provided by SageMaker Data Wrangler.
Get started with the workshop »
- SageMaker Ground Truth
-
-
Getting Started with Amazon Ground Truth
Explore how to use the console to create a labeling job, assign a public or private workforce, and send the labeling job to your workforce. Learn how to monitor the progress of a labeling job.
Explore the guide »
-
Label Training Data for Machine Learning
Learn how to set up a labeling job in Amazon SageMaker Ground Truth to annotate training data for your ML model.
Get started with the tutorial »
-
Getting started with Amazon Ground Truth Plus
Explore how to complete the necessary steps to start an Amazon SageMaker Ground Truth Plus project, review labels, and satisfy SageMaker Ground Truth Plus prerequisites.
Explore the guide »
-
Get started with Amazon Ground Truth
Watch how to get started with labeling your data in minutes through the SageMaker Ground Truth console.
Watch the video »
-
Amazon SageMaker Ground Truth Plus – create training datasets without code or in-house resources
Learn about Ground Truth Plus, a turn-key service that uses an expert workforce to deliver high-quality training datasets fast, and reduces costs by up to 40 percent.
Read the blog »
- SageMaker JumpStart
-
-
Get started with machine learning with SageMaker JumpStart
Explore SageMaker JumpStart solution templates that set up infrastructure for common use cases, and executable example notebooks for machine learning with SageMaker.
Explore the guide »
-
Get Started with your machine learning project quickly using Amazon SageMaker JumpStart
Learn how to fast-track your ML project using pretrained models and prebuilt solutions offered by Amazon SageMaker JumpStart. You can then deploy the selected model through Amazon SageMaker Studio notebooks.
Get started with the tutorial »
-
Get hands-on with Amazon SageMaker JumpStart with this Immersion Day workshop
Learn how the low-code ML capabilities found in Amazon SageMaker Data Wrangler, Autopilot and Jumpstart, make it easier to experiment faster and bring highly accurate models to production.
Get started with the workshop »
- SageMaker Pipelines
-
-
Getting Started with Amazon SageMaker Pipelines
Learn how to create end-to-end workflows that manage and deploy SageMaker jobs. SageMaker Pipelines comes with SageMaker Python SDK integration, so you can build each step of your pipeline using a Python-based interface.
Explore the guide »
-
Automate machine learning workflows
Learn how to create and automate end-to-end machine learning (ML) workflows using Amazon SageMaker Pipelines, Amazon SageMaker Model Registry, and Amazon SageMaker Clarify.
Get started with the tutorial »
-
How to create fully automated ML workflows with Amazon SageMaker Pipelines
Learn about Amazon SageMaker Pipelines, the world’s first ML CI/CD service designed to be accessible for every developer and data scientist. SageMaker Pipelines brings CI/CD pipelines to ML, reducing the coding time required.
Watch the video »
- SageMaker Studio
-
-
Build and train a machine learning model locally
Learn how to build and train a ML model locally within your Amazon SageMaker Studio notebook.
Get started with the tutorial »
-
SageMaker Studio integration with EMR workshop
Learn how to utilize distributed processing at scale to prepare data and subsequently train machine learning models.
Get started with the workshop »
The third set of AI/ML AWS service pathways provide in-depth
documentation, hands-on tutorials, and resources to get started with
AWS Trainium, AWS Inferentia, and Amazon Titan.
- AWS Trainium
-
-
Scaling distributed training with AWS Trainium and Amazon EKS
Learn how you can benefit from the general availability of Amazon EC2 Trn1 instances powered by AWS Trainium—a purpose-built ML accelerator optimized to provide a high-performance, cost-effective, and massively scalable platform for training deep learning models in the cloud.
Read the blog »
-
Overview of AWS Trainium
Learn about AWS Trainium, the second-generation machine learning (ML) accelerator that AWS purpose built for deep learning training of 100B+ parameter models. Each Amazon Elastic Compute Cloud (EC2) Trn1 instance deploys up to 16 AWS Trainium accelerators to deliver a high-performance, low-cost solution for deep learning (DL) training in the cloud.
Explore the guide »
-
Recommended Trainium Instances
Explore how AWS Trainium instances are designed to provide high performance and cost efficiency for deep learning model inference workloads.
Explore the guide »
- AWS Inferentia
-
-
Overview of AWS Inferentia
Understand how accelerators are designed by AWS to deliver high performance at the lowest cost for your deep learning (DL) inference applications.
Explore the guide »
-
AWS Inferentia2 builds on AWS Inferentia1 by delivering 4x higher throughput and 10x lower latency
Understand what AWS Inferentia2 is optimized for - and explores how it was designed from the ground up to deliver higher performance while lowering the cost of LLMs and generative AI inference.
Read the blog »
-
Machine learning inference using AWS Inferentia
Learn how to create an Amazon EKS cluster with nodes running Amazon EC2 Inf1 instances and (optionally) deploy a sample application. Amazon EC2 Inf1 instances are powered by AWS Inferentia chips, which are custom built by AWS to provide high performance and lowest cost inference in the cloud.
Explore the guide »
- Amazon Titan
-
-
Overview of Amazon Titan
Explore how Amazon Titan FMs are pretrained on large datasets, making them powerful, general-purpose models. Learn how you can use them as is - or privately - to customize them with your own data for a particular task without annotating large volumes of data.
Explore the guide »
Explore
-
Architecture diagrams
These reference architecture diagrams show examples of AWS AI and ML services in use.
Explore architecture diagrams »
-
Whitepapers
Explore whitepapers to help you get started and learn best practices in choosing and using AI/ML services.
Explore whitepapers »
-
AWS Solutions
Explore vetted solutions and architectural guidance for common use cases for AI and ML services.
Explore solutions »
Resources
Foundation models
Supported foundation models include:
Using Amazon Bedrock, you can experiment with a variety of foundation
models and privately customize them with your data.
Use case or industry-specific
services
Associated blog posts
