What is machine learning as a service (MLaaS)?

In this post, we’ll be covering some of the basics of machine learning and machine learning as a service (MLaaS) and comparing how the ML services of Microsoft Azure, Amazon Web Services (AWS), and Google Cloud Platform (GCP) stack up.

Along with serverless, artificial intelligence (AI) and machine learning (ML) might just be the killer app for the cloud, combining massive data handling with virtually limitless computing power and pay-only-for-what-you-need economic model.

But what is machine learning, and how does machine learning as a service work? Read on!

• What is machine learning as a service (MLaaS)?
• AI vs ML: What’s the difference between machine learning and artificial intelligence?
• Benefits (and limitations) of machine learning
• Machine learning cloud services compared
• How to learn AI and ML

Get the Cloud Dictionary of Pain
Speaking cloud doesn’t have to be hard. We analyzed millions of responses to ID the top concepts that trip people up. Grab this cloud guide for succinct definitions of some of the most painful cloud terms.

Machine learning as a service (or MLaaS) refers to the wide range of machine learning tools offered as services from cloud computing providers.

Similar to cloud service models such as SaaS (software as a service) or PaaS (platform as a service), using machine learning as a service means getting instant access to powerful tools over the internet without the money or expertise needed to create them yourself.

That last part is kind of a big deal. 

Not too long ago, the equipment and expertise needed to do anything close to machine learning or artificial intelligence was so prohibitively expensive and specialized that only governments and a few universities could afford it.

In a broad sense, you need to have three things to get those machines a-learnin’.

• Data — Lots and lots of it
• Computation — Some way to apply computation or algorithms to that data
• Knowledge — You kind of need to know what you're doing.

With machine learning as a service, cloud computing has managed to bring these three things within reach of anyone who happens to have an internet connection.

Today, we can manage massive amounts of data and harness immense cloud computing power using point-and-click tools that the cloud providers have created and only pay for specifically what we need.

While knowledge is still important, cloud providers have created some turnkey services that let us make use of very powerful machine learning technology through a simple API call.

Artificial intelligence and machine learning are often used interchangeably by the popular press, but AI and machine learning are NOT the same thing — at least in the eyes of the AI community.

• Artificial intelligence is our pursuit of simulating human thought and decision in an automated fashion.

• Machine learning is — at least according to Arthur Samuel, the guy who coined the term back in 1959 — “the field of study that gives computers the ability to learn without being explicitly programmed.”

In other words, machine learning is one method we can use to try to achieve artificial intelligence. That said, even some cloud providers make liberal use of the term AI and ML. So just be aware of this as it's usually part marketing-speak.

The benefits of machine learning for businesses are massive. Machine learning can automate tedious and time-consuming manual processes, more efficiently handle data, reduce human errors, and help drive continuous improvement.

Machine learning can be used for everything from business forecasting to spam detection to improved customer services.

Did you know that a Machine-Learning Engineer is one of the most in-demand cloud professions of 2021? Check out other top-paying cloud certifications and jobs!

Many articles on ML read like “ML is great! Would you like a slice?” But there are some limitations to machine learning.

For all its promise and opportunity, developing quality machine learning models is really hard. If you get it wrong, the resulting ML-generated decisions can range anywhere from slightly embarrassing to downright immoral.

There’s also the fact that if your source data is weak, your results may be disappointing.

Both for ethical and sometimes regulatory reasons, we need to be able to explain how our machine learning model makes its decisions. Practitioners call this explainability. Fortunately, our cloud providers have tools to help us out in this area.

• AWS has SageMaker Clarify, which can help provide a view into how data elements influence the model generation process and evaluate fairness.
• Azure has this ability integrated into Responsible ML and Fairness SDK.
• GCP provides this under the name AI Explanations. 

Let’s get into comparing AWS machine learning services with Azure machine learning services and GCP machine learning services.

We’ll look at the ML offerings of AWS vs GCP vs Azure across three different areas:

• Machine learning platforms
• Machine learning infrastructure
• Machine learning building blocks

Disclaimer: Cloud providers are investing massive amounts of resources into expanding their machine learning offerings, so features and services are going to be always evolving, at least for the foreseeable future.

This info can hopefully give you some background on services offered and the respective terminology used by cloud providers, but don't expect some declaration about one provider being better than the others. This space is just too volatile.

We'll get started with machine learning building blocks, as these are usually the most common way people get started with machine learning because the barrier to entry is so low.

These are ready-to-go services that are available as an API call or using the SDK from the cloud provider. Now all the providers we're going to talk about here offer rest APIs for their machine learning services.

Let's take a look at some common ML uses and see what we have. 

Text-to-speech and speech-to-text services are cloud services for converting text to audible speech and vice versa.

Speech to text and text to speech are things that we probably use daily and maybe take for granted, but there are some pretty complex things going on behind the scenes. Fortunately, our cloud providers have us covered. 

As you can see, Amazon gets a bit more out there with their naming, conjures up images of a parrot by calling their service Amazon Polly, while Azure and GCP go with more obvious naming of Speech to Text.

Like it or not, chatbots have started becoming more commonplace as a first line of customer support. Our cloud providers are doing their part to help chatbots be less disappointing by creating services. 

If you’ve been around the internet since back in the day, you might recall a website called Babelfish. Babelfish was a free language translation website and, for the late ’90s, I thought it was just about the most amazing slice of technology I had ever seen.

Of course, language translation is more table stakes than novelty these days. Fortunately, all three of our cloud providers have chosen to stick to literal naming for their own translation service.

Text analytics services can take natural language, meaning how we speak to one another, and extract certain themes, topics, and sentiments. 

An evolution of text analytics is document analysis. Document analysis is where machine learning can do stuff like summarize articles or detect information in forms.

Image analysis and video analysis services can recognize objects and people in pictures, map faces, or detect potentially objectionable content. 

Computers are pretty good at detecting when things are out of the ordinary, but you normally have to tell them specifically what to watch. Cloud providers have used machine learning to create services that can just watch a stream of events or data and figure out what's different. This is called anomaly detection.  

Recommendation engines are becoming a popular addition to e-commerce sites, and our cloud providers have tried to do the heavy lifting for us here.

Now, one thing to keep in mind here is your recommendations will only be as good as the transactional data you're able to feed in. In fact, that goes for most all these services. If your source data is sketchy, your results may be disappointing.

Automating AWS Cost Optimization
Using AWS cost-effectively can be a challenge. In this free, on-demand webinar, you'll get an overview of AWS cost-optimization tools and strategies, like data storage optimization.

When I talk about machine learning platforms, I'm referring to the workbench and tools that ML practitioners would use. It's analogous to a developer using an IDE and some libraries to write their code.

For machine learning, Jupyter Notebook is the current de facto workbench for data scientists, so it's no surprise that all the cloud providers offer Jupyter Notebooks or some slightly rebranded version as part of their platforms.

Another consistency is in the support of major machine learning frameworks TensorFlow, MXNet, Keras, PyTorch, Chainer, SciKit Learn, and several more are fully supported.

Features such as security, collaboration, and data management are all well integrated by all the vendors, but the specifics on how you use these things varies by provider.

For those who are just starting out on their machine learning journey, our cloud providers have invested in some gentle introductions.

• AWS calls their “just getting started” service SageMaker Autopilot.
• Azure has Automated ML and a neat drag-and-drop tool called Designer.
• GCP has a line of guided model creation tools that they call AutoML.

For the seasoned pro who doesn't need the training wheels:

• AWS offers SageMaker Studio.
• Azure has Machine Learning Notebook.
• GCP calls their main ML development platform simply AI Platform. 

Another feature getting lots of attention as of late is the DevOps equivalent for machine learning, so-called MLOps.

• Azure just calls their MLOps offering MLOps.
• AWS has SageMaker MLOps.
• GCP accomplishes this via their Pipeline service. 

AWS has Augmented AI, something that I haven't seen on the other platforms yet, but I'm sure that's just a matter of time.

Augmented AI is a way to enlist the reasoning power of teams of real live humans to help improve your machine learning service.

Let's say you've determined that your machine learning model is about 95% accurate at identifying pictures of angry ferrets. But you must have 100% accuracy. For those cases where the ML model's confidence is low, you can direct that picture over to a live human, which will then make the determination of “anger” or “not anger.”

All of our cloud providers really, really like containers for their respective machine learning platforms. And this is for good reason. Containers are relatively lightweight, portable, can be shuffled around without much hassle.

All the providers offer push-button deployment of containers for specific versions of the ML frameworks, optimized for training validation and inferences.

If you're more of a do-it-yourself person, all the providers have platform-optimized virtual machines for all the major frameworks as well. Most people use this option if they already have a model trained on-prem. For example, if you already have a model created using PyTorch, you can just spin up a VM with that specific version of PyTorch in the cloud and copy your model out there. 

Machine learning Hardware

There is a bit of an arms race with machine learning optimized hardware among the cloud providers, each claiming superior performance and economics.

All of the providers offer various levels of CPU and GPU virtual machine types.

Additionally, some have also invested in specialized hardware in the form of application-specific integrated circuits and field-programmable gate arrays. 

• AWS offers Habana Gaudi ASIC instances and a custom processor they call AWS Trainium optimized for model training. They also offer an ASIC called Inferentia for machine learning inferences.

• GCP has long offered their custom tensor processing unit (or TPU), which is an ASIC optimized for the TensorFlow framework.

• Not to be outdone, Azure offers a line of FPGA-based virtual machines tuned specifically for machine learning workloads.

Now there is a tradeoff here. These specialized hardware platforms are really good at machine learning tasks, but they're not much good for anything else. Economically, CPU- and GPU-based machines are much more flexible and generally what people use first, as they develop and refine their ML models.

Machine learning is a rapidly evolving and iterating space and the cloud has just accelerated that even more. 

If you're just getting started on your ML journey, check out the Intro to Machine Learning course on ACG by yours truly. I use history, simile, and illustrations to demystify all those scary words.

Once you have the basics, you can then pick your cloud and dive a bit deeper. We have courses and hands-on labs to let you dive deep into the ML offerings of AWS, GCP, and Azure.

• AWS Certified Machine Learning - Specialty
• Getting Started with Azure Machine Learning Studio
• Google Cloud AI Services Deep Dive

Not sure where to begin? Check out ACG Learning Paths that will teach you AI, ML, and data skills for AWS, Azure, and GCP.

• AWS Machine Learning certification path
• Microsoft Azure Data learning path
• GCP ML learning path

Thanks for reading. And keep being awesome, cloud gurus!

Master modern tech skills, get certified, and level up your career. Whether you’re starting out or a seasoned pro, you can learn by doing and advance your career in cloud with ACG.