ML Cloud Platforms Comparison: IBM Watson, Google Cloud, AWS, Azure

Here's a detailed breakdown of all four platforms to help you decide which option matches your business needs best.

1. Amazon AWS

With over a million of active users, Amazon Web Services (AWS) is easily the biggest provider of cloud-based services today. No wonder that AWS jumped on the machine learning bandwagon early on and launched the Amazon Machine Learning service already in 2015. Today, the platform features several other services to help developers create models and put data to work.

Pros

• Supports all major machine learning frameworks
• Offers comprehensive data analytics services and powerful predictive capabilities
• Equips developer teams with tools, APIs, and software development kits (SDKs) designed to make developing predictive applications easier
• Includes an automatic data transformation tool
• Ensures security with granular permission policy
• Praised for scalability

Cons

• Many people consider AWS billing confusing
• You need to store your data in AWS before using it in the Machine Learning service, which incurs extra fees
• Steep learning curve and high complexity barrier
• AWS doesn't come with default enterprise-grade support – customers need to buy the Business tier support, which might mean a premium of up to 10 percent on the overall AWS spending
• Suffered from some significant outages during the last three years

For whom?

AWS is a solution for large enterprises that plan to take advantage of its many services and features. Only then it becomes cost-effective to upgrade the platform. AWS isn't the best option for small and mid-sized businesses, as they often get less support and guidance from AWS representatives than larger organisations.

2. Microsoft Azure

Microsoft Azure has gained a lot of traction in the cloud storage and business scenes recently. And its service range is continuously expanding – one recently added service is the Azure Machine Learning Studio that allows developers to write, test, and deploy machine learning algorithms. The platform offers a long list of predefined algorithms and a marketplace for off-the-shelf APIs to help kick off your data projects.

Pros

• Incredibly high availability and redundancy in global data centers, offering service level agreements at 99.95% (that means c. 4.38 hours of downtime per year)
• Easily scalable computing power to give businesses the flexibility they need
• Cost-effectiveness: thanks to pay-as-you-go pricing, Azure helps organisations manage their IT budgets
• Security: Azure combines the standard security model of Detect, Assess, Diagnose, Stabilise, and Close with strong cybersecurity controls to deliver multi-level protection to end users and business data

Cons

• It needs to be managed and maintained by experts – since your computing power moves to the cloud, you need to have people on board who will be able to manage it (this includes monitoring servers and applying patches)
• It offers less automation than other vendors, which might impact your developer team's productivity

For whom?

Azure is an excellent option if your organisation is already invested in Microsoft's technology and developer skills. Companies which have that expertise will be able to take full advantage of Azure and ensure that services aren't over-provisioned, preventing Azure costs from skyrocketing.

3. Google Cloud

Google's Cloud Machine Learning Engine helps developers build models on the basis of its open source TensorFlow library and run predictions at scale without having to worry about the infrastructure. Google provides users with its state-of-the-art algorithms used in search, as well as plenty off-the-shelf APIs for such features as natural language processing, translation, and computer vision.

Pros

• Based on Google’s public cloud infrastructure, which helps to drive down costs, boosts accessibility, and reduces requirements for infrastructure expertise
• Continued usage of Google Cloud is cheaper than other platforms
• Integration with other Google services

Cons

• There's no reliable support available from Google
• You need to store data and stage models in Google Cloud Storage
• Features minimal abstraction which gives the power to developers, but non-technical staff will encounter a steep learning curve
• Transitioning away from Google Cloud Platform is challenging

For whom?

Google Cloud Platform offers particular strengths in data analytics that will make it an appealing choice for organisations planning to launch products in areas such as the Internet of Things.

4. IBM Watson

A mature cognitive computing platform, IBM Watson offers a powerful data ingestion engine and machine learning-as-a-service. The machine learning component is focused on helping developers in getting their models to production.

Pros

• Developers can do all their work in one place
• Access to over 30 types of data stores that are part of the Watson Data Platform
• Community support: developers can share sets, notebooks, articles and other resources for easier learning

Cons

• The service is limited, because it's geared towards building machine learning-based apps through API connections

For whom?

IBM Watson’s cloud-based predictive analytics and cognitive computing services come in handy for large enterprises and currentlypower various business initiatives of organisations such as General Motors, Condé Nast, GlaxoSmithKline or the American Cancer Society.

Key takeaway

According to Forrester, the market for machine learning platforms will grow at the rate of 15% a year through 2021. That's why it's likely that we'll see more machine learning-oriented expansions in these platforms and some interesting newcomers like Baidu.

Have you got any questions about machine learning platforms? Or perhaps you're already using one of these solutions? Leave a comment and share your experience to help the community learn more about industry best practices in the cutting-edge area of machine learning.