We use essential cookies and similar tools that are necessary to provide our site and services. We use performance cookies to collect anonymous statistics, so we can understand how customers use our site and make improvements. Essential cookies cannot be deactivated, but you can choose “Customize” or “Decline” to decline performance cookies.
If you agree, AWS and approved third parties will also use cookies to provide useful site features, remember your preferences, and display relevant content, including relevant advertising. To accept or decline all non-essential cookies, choose “Accept” or “Decline.” To make more detailed choices, choose “Customize.”
Customize cookie preferences
We use cookies and similar tools (collectively, "cookies") for the following purposes.
Essential
Essential cookies are necessary to provide our site and services and cannot be deactivated. They are usually set in response to your actions on the site, such as setting your privacy preferences, signing in, or filling in forms.
Performance
Performance cookies provide anonymous statistics about how customers navigate our site so we can improve site experience and performance. Approved third parties may perform analytics on our behalf, but they cannot use the data for their own purposes.
Allowed
Functional
Functional cookies help us provide useful site features, remember your preferences, and display relevant content. Approved third parties may set these cookies to provide certain site features. If you do not allow these cookies, then some or all of these services may not function properly.
Allowed
Advertising
Advertising cookies may be set through our site by us or our advertising partners and help us deliver relevant marketing content. If you do not allow these cookies, you will experience less relevant advertising.
Allowed
Blocking some types of cookies may impact your experience of our sites. You may review and change your choices at any time by selecting Cookie preferences in the footer of this site. We and selected third-parties use cookies or similar technologies as specified in the AWS Cookie Notice.
Unable to save cookie preferences
We will only store essential cookies at this time, because we were unable to save your cookie preferences.
If you want to change your cookie preferences, try again later using the link in the AWS console footer, or contact support if the problem persists.
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 AI
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 AI 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.
In this 7 minute video excerpt, Rajneesh Singh, general manager of Amazon SageMaker AI Low-Code/No-Code
team at AWS, explains how machine learning can address business
problems.
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.
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 AI 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 AI: This service provides the most flexibility and
control over the ML algorithm. You can use SageMaker AI 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 AI: SageMaker AI 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 AI. 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 AI, 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 AI, 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.
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 AI 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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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 AI family.
SageMaker AI
How Amazon SageMaker AI works
Explore the overview of machine learning and how SageMaker AI works.
Explore how Amazon SageMaker AI 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 modeling with Amazon SageMaker AI 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 AI Autopilot.
Generate machine learning predictions without writing code
This tutorial explains how to use Amazon SageMaker AI Canvas to build ML models and generate accurate predictions without writing a single line of code.
Use Amazon SageMaker AI Canvas to make your first ML Model
Learn how to use Amazon SageMaker AI Canvas to create an ML model to assess customer retention, based on an email campaign for new products and services.
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 AI Data Wrangler.
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.
Getting started with Amazon Ground Truth Plus
Explore how to complete the necessary steps to start an Amazon SageMaker AI Ground Truth Plus project, review labels, and satisfy SageMaker AI Ground Truth Plus prerequisites.
Amazon SageMaker AI 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.
Get started with machine learning with SageMaker AI JumpStart
Explore SageMaker AI JumpStart solution templates that set up infrastructure for common use cases, and executable example notebooks for machine learning with SageMaker AI.
Get Started with your machine learning project quickly using Amazon SageMaker AI JumpStart
Learn how to fast-track your ML project using pretrained models and prebuilt solutions offered by Amazon SageMaker AI JumpStart. You can then deploy the selected model through Amazon SageMaker AI Studio notebooks.
Get hands-on with Amazon SageMaker AI JumpStart with this Immersion Day workshop
Learn how the low-code ML capabilities found in Amazon SageMaker AI Data Wrangler, Autopilot and Jumpstart, make it easier to experiment faster and bring highly accurate models to production.
Getting Started with Amazon SageMaker AI Pipelines
Learn how to create end-to-end workflows that manage and deploy SageMaker AI jobs. SageMaker AI Pipelines comes with SageMaker AI Python SDK integration, so you can build each step of your pipeline using a Python-based interface.
Learn how to create and automate end-to-end machine learning (ML) workflows using Amazon SageMaker AI Pipelines, Amazon SageMaker AI Model Registry, and Amazon SageMaker AI Clarify.
How to create fully automated ML workflows with Amazon SageMaker AI Pipelines
Learn about Amazon SageMaker AI Pipelines, the world’s first ML CI/CD service designed to be accessible for every developer and data scientist. SageMaker AI Pipelines brings CI/CD pipelines to ML, reducing the coding time required.
Explore how Amazon SageMaker AI 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.
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.
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.
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.
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 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.
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.
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.