Machine Learning services in the cloud are a critical area of the modern computing landscape, providing a way for organizations to better analyze data and derive new insights. Accessing these service via the cloud tends to be efficient in terms of cost and staff hours.
Machine Learning (often abbreviated as ML) is a subset of Artificial Intelligence (AI) and attempts to ‘learn’ from data sets in several different ways, including both supervised and unsupervised learning. There are many different technologies that can be used for machine learning, with a variety of commercial tools as well as open source framework.
While organizations can choose to deploy machine learning frameworks on premises, it is typically a complex and resource intensive exercise. Machine Learning benefits from specialized hardware including inference chips and optimized GPUs. Machine Learning frameworks can also often be challenging to deploy and configure properly. Complexity has led to the rise of Machine Learning services in the cloud, that provide the right hardware and optimally configured software to that enable organizations to easily get started with Machine Learning.
There are several key features that are part of most machine learning cloud services.
AutoML – The automated Machine Learning feature automatically helps to build the right model.
Machine Learning Studio – The studio concept is all about providing a developer environment where machine learning models and data modelling scenarios can be built.
Open source framework support – The ability to support an existing framework such as TensorFlow, MXNet and Caffe is important as it helps to enable model portability.
How To Choose
When evaluating the different options for machine learning services in the cloud, consider the following criteria:
- Existing usage – Each of the major public cloud providers has its own machine learning service. It’s often an easy first option to simply stick with the same platform where your data already resides.
- Data access – The ability to ingest data, or pull data sets from any required source is an important consideration, otherwise a lot of time will be wasted simply moving data around.
- Workflow modelling – Machine Learning can be a complex activity, which is why it’s a good idea to make sure that there is a workflow modelling capability in place that is easy to work with.
In this Datamation top companies list, we spotlight the vendors that offer the top machine learning services in the cloud.
- Amazon Web Services
- Google Cloud
- IBM Watson Machine Learning
- Microsoft Azure
- Salesforce Einstein
Alibaba Machine Learning Platform for AI (PAI)
Value proposition for potential buyers: Alibaba is a great option for users that have machine learning needs where data sets reside around the world and especially in Asia, where Alibaba is a leading cloud service.
- A key differentiator for Alibaba is its PAI Studio tool which integrates pre-built modules for data preprocessing, feature engineering and statistical analysis
- Automatic parameter tuning with AutoML is powerful feature that helps users to automatically fine tune an algorithm to get the desired results.
- Visual interface help users to set up ML workflow in a drag and drop approach.
- Supports multiple common ML frameworks including TensorFlow, MXNet and Caffe
Amazon Web Services (AWS)
Value proposition for potential buyers: Amazon Web Services has the broadest array of machine learning services in the cloud today, leading with its SageMaker portfolio that includes capabilities for building, training and deploying models in the cloud.
- SageMaker is a fully managed offering from AWS that has multiple services including:
- Ground Truth – for building and managing training data sets
- Studio – a full integrated development environment for ML
- Autopilot – for automatically building and training models
- Model Tuning – for paramater optimization
- A key differentiator for AWS is the extensibility of the SageMaker services with Notebooks that lets users share and collaborate on ML models. Additionally the AWS marketplace offers third party pre-built algorithms and models for users to consume.
- Supported frameworks include TensorFlow, PyTorch, Apache MXNet, Chainer, Keras, Gluon, Horovod, Scikit-learn, and Deep Graph Library.
Value proposition for potential buyers: Google’s set of Machine Learning services are also expansive and growing, with both generic as well as purpose built services for specific use-cases.
- The core element of Google’s machine learning services in the cloud is the Cloud AutoML suite which is designed to be an easy on ramp to help users get started building and deploying models.
- Making ML easier is also the goal of the AI hub which has a repository of component developers can use to build models.
- An AI or ML model is only as good as the data that it is based on, which is why the AI Platform Data Labeling Service from Google is so useful, helping to properly prepare and identify the right data for machine learning.
- A key differentiator is Vision AI and Video AI which are focused tools for video and audio data.
Value proposition for potential buyers: IBM Watson Machine learning enables users to run models on any cloud, or just on the the IBM Cloud
- A key differentiator for IBM Watson machine learning is the local component that enables users to first build a model on-premises and then use that same model to run on any cloud.
- On IBM Cloud, Watson Machine Learning is fully managed, including the Watson Studio developer environment for building and deploying models.
- It’s not enough to just run model, it’s also important to monitor and measure outcomes, which is where the IBM Watson OpenScale service fits in with IBM Cloud, providing a governance and monitoring model for AI.
- GPU accelerated machine learning training on IBM Cloud supports the Keras, PyTorch, Tensorflow and Caffe frameworks.
Value proposition for potential buyers: For organizations that have already bought into Microsoft Azure cloud, Azure Machine Learning is good fit, providing a cloud environment to train, deploy and manage machine learning models.
- A key differentiator for Azure Machine Learning is the service’s intuitive drag and drop designer for building machine learning models
- Microsoft has embraced the concept of MLOps for its platform providing a DevOps-style approach for building and managing machine learning pipelines and workflows.
- Another area of strength for Azure Machine Learning is the services integrated security and governance controls that can help align machine learning efforts with compliance effort as well as identity and privacy controls.
- Support for multiple open source frameworks including PyTorch, TensorFlow, Kera and scikit-learn.
Value proposition for potential buyers: Oracle Machine learning is a useful tools for organizations already using Oracle Cloud applications, to help build data mining notebooks.
- Oracle Machine Learning (OML) includes several services such as OML Notebooks, OML Microservices, OML4SQL as well as the Oracle Data Miner.
- A key feature of the Oracle Machine Learning services is the integrated collaboration capabilities that help users work together.
- OML services are well integrated with Oracle applications including the company’s namesake database
Value proposition for potential buyers: Salesforce Einstein is a purpose built machine learning platform that is tightly integrated with the Salesforce platform.
- Salesforce Einstein has a primary purpose of helping Salesforce application users to get better insights from their own data.
- Beyond just working with existing Salesforce applications, Einstein can be used to build artificial intelligence powered apps that are delivered from the Saleforce application cloud.
- Einstein Discovery is the core machine learning service that can also be used to find insights and patterns in data that doesn’t reside within Salesforce.