What is Machine Learning as a Service?
MLaaS refers to the different cloud-based platforms which handle most infrastructural issues related to machine learning. These include model training, model evaluation and data pre-processing among other things.
Some of the leading MLaaS providers in the market include Microsoft Azure Machine Learning, Amazon Machine Learning services and Google Cloud AI. All of these services facilitate fast training of models as well as deployment.
According to a research report by Gartner, Amazon had the greatest market share in 2017 with 51.8%. Microsoft was in second place while Google took the third position. However, in terms of revenue, Microsoft took the lead in the same year, with an annual run rate of $21.2 billion. Amazon had $20.4 billion while Google was way behind the two, with only $4.4 billion.
Amazon and Microsoft are clearly the top dogs in enterprise adoption of cloud services such as MLaaS while Google still lags behind. Let us start by considering the different machine-learning-as-a-service platforms that the three provide to understand the disparities.
Amazon Web Services (AWS)
Amazon Web Services offers two main services for businesses
a) Predictive Analytics
In the current market, AWS predictive analytics is the tool with the highest level of automation. Moreover, it is also one of the best for operations with strict deadlines. It is capable of loading data from a number of sources including CSV files, Amazon Redshift and Amazon RDS among others.
The service performs all pre-processing automatically, identifying numerical and categorical fields. In all of these processes, it does not give the user an opportunity to select the methods they would like to use for further pre-processing if necessary.
This tool limits prediction capacities to three main alternatives:
- Binary classification
- Multiclass classification
The main benefit of this tool is the fact that it does not require users to know any machine learning as everything is automated. But it does not offer support for any unsupervised learning approaches. All in all, it is a great solution if you are looking for fully automated service and do not mind the limitations.
b) SageMaker Tool (for data scientists)
An alternative to this is SageMaker, a service that aims to provide data scientists with the tools they need to build models and deploy them fast. For example, it includes built-in algorithms optimized for computations in distributed frameworks and large datasets. It also allows data scientists to simplify data analysis and exploration without the hassle of server management.
Data scientists can use the APIs that the company provides or get a little more creative and use their own data sets.
Overall, AWS is flexible enough to accommodate the varying needs of expert data scientists as well as those without much experience.
Azure has the ultimate objective of creating the ideal framework for both experienced data scientists and newcomers in the field, just like Amazon. But what sets apart the two is the fact that Azure offers greater flexibility when it comes to ready-to-use algorithms.
a) Azure Machine Learning Studio
The main tool, Azure Machine Learning Studio, makes use of a graphical drag-and-drop interface for almost all processes. Even though using the tool requires some learning curve, it eventually equips users with basic knowledge of how machine learning works. To support newcomers, it offers a visual peek into every step of the workflow.
Its top advantage is that it offers support for over 100 algorithms that handle:
- Binary and Multi-class Classification
- Text analysis
- Anomaly detection
Another major merit is the Cortana Intelligence Gallery, a collection of machine learning tools that the Microsoft community provides for exploration and reuse.
b) Microsoft Azure Machine Learning Service
Slightly different from Machine Learning Studio, this tool is designed for experienced data scientists. It does not have built-in out-of-box solutions but rather, requires custom engineering. The tool incorporates Python packages with libraries and functions for various tasks and visual studio tools for AI among other things.
Though it has a lot more capabilities, using this option requires a powerful team of data scientists as well as AI developers.
Google Cloud AI
Previously, Google used to have two AI service levels, a Google Prediction API (highly automated) and a machine learning engine for experienced developers.
a) Google Cloud AutoML
Unfortunately though, they pulled the plug on the Prediction API. In its place, Google is testing Cloud AutoML which is similarly aimed at inexperienced users looking to deploy machine learning models. However, it is yet to launch.
b) Google Cloud Machine Learning Engine
This is one of the most flexible platforms for data scientists. It recommends the use of TensorFlow and Google Cloud infrastructure as its drivers. TensorFlow is a Google product which is essentially an open source library of various machine learning tools. It has a rather steep learning curve and no visual interface. Its main targets are software engineers who are looking to transition into data science.
The service claims to offer user-friendly approaches for building machine learning models for any type and volume of data. Furthermore, the service is integrated with all Google services and has a focus on deep neural networks.
Top 5 Factors Businesses Consider Before Choosing an ML Service
The above three service providers are just a tip of the iceberg when it comes to MLaaS. Searching for the right provider to suit your business needs can seem complex due to the high number of vendors and offerings available.
Many companies make this process easier by going through a checklist of factors to ensure that they make the most gain from the purchase. Let us go through five of the most important factors that one should weigh:
1. Services and Features
The top priority for any company is to find a suite of features and services that address its wants and needs. In some cases, achieving this requires that an organization makes use of several providers, each targeting a different aspect of operations.
In the case of Amazon, Microsoft and Google, the capabilities are by and large similar, with flexibility in computing, storage and networking requirements. However, as mentioned above, Google discontinued its basic user package and is in the process of launching another.
Amazon, on the other hand, is renowned for its wide range of MLaaS services that cover almost every facet of the business as it has been around the longest.
One of the most important factors to look for in this highly dynamic field is a partner who keeps growing and innovating to offer relevant services at all times. How often do they update their software? What servers are they using? Are they using the latest technology in machine learning?
Generally speaking, businesses seek service providers who are flexible in terms of change and innovation.
3. Technical Capabilities
For a service to yield any amount of benefits, a company should be in a position to implement it smoothly and efficiently. Consider the learning curve of any given service and the implementation process.
If you already have a team of data scientists or the budgetary resources to get one, this may not be much of a problem. But in case you want to find a way to implement a simple solution without extra manpower, consider the automated solutions offered by the different providers.
Check the feasibility of a program against your company’s circumstances and needs before jumping on board.
Pricing is a huge factor to consider when seeking an MLaaS solution. For all the three top competing providers, high competition has created a sustained downward trend. Though their prices are not very different, each one offers a different pricing model, discount framework and periodic price cuts.
How does a given solution work? Will it disrupt my current workflow or enhance it? Successful implementation of machine learning usually requires proper identification of the problem at hand and how the technology can solve it.
By the time a company gets to the point of selecting vendors, they need to know exactly what problems they are looking to solve and the benefits they expect. With this in mind, weigh the functionality of a given system against business needs before making a pick.
What the Future Holds for Machine Learning as a Service
As more companies realize the massive potential machine learning can ring to their businesses, the machine-learning-as-a-service market seems set to keep growing. According to a report by Research and Markets, it is expected to register a CAGR of 49% during the period between 2017 and 2023.
With MLaaS, businesses can reap the benefits of the technology without undertaking the rigorous process of building their own models from scratch. They no longer have to worry so much about in-house capability or having data science and engineering degrees. Rather, the main focus is on getting the solution that best integrates into their existing framework and delivers the most value.
Back to insights