Machine Learning as a Service – A Closer Look at Azure and AWS

 

Machine learning is an application of AI where outcomes are predicted in advance, and the technology effectively learns and improves over time. This dynamic occurs without the need for human intervention, as machine learning is able to process data and pull insights on its own. Machine learning as a service (MLaaS) is a broader term of cloud platforms that can automate various processes. This can include data processing and evaluating models along with outputting predictions. Two of the leading platforms for MLaaS are Microsoft Azure and Machine Learning on Amazon Web Services. Each offers users speedy modeling training and easy deployments without the need for extensive data science experience. And both platforms hold the promise of helping firms to see their data as a way to look predictively towards (likely) future events, not just analyze what has already occurred.

 

Machine learning applications are moving into various industry applications. Google is utilizing machine learning to predict flight delays based on data points including location, weather, and late aircraft arrivals. The tool will compile this data and enable the flight to appear as “delayed” on booking and status engines when it reaches a certain likelihood threshold.

 

Microsoft’s Azure platform is used by UK-based Callcredit to determine if borrowers are at a higher risk of default. Callcredit utilizes Azure’s capabilities to predict problems with credit rating assessments and predictively spot fraudulent applications. It offered this enhanced machine learning to its customers such as credit card companies to help them avoid millions in bad debt. It’s also used by North American Eagle in their bid to break the land speed record. They’re using Azure Machine Learning to process data sets about speed performance in completely new ways in record time. The group uses this data gleaned from prior and current speed runs to build predictive models to help them increase speed while ensuring the safety of the human driver.

 

AWS’ platform is built with more automation and accessibility so that it appeals to a broader group of individuals who might not possess data science skills. With Azure’s Machine Learning, there is an assumption that the user understands modeling and the algorithms, but appreciates a more intuitive and friendly GUI.

 

Machine Learning for “The Masses”

 

Both of the platforms are seen as a broader “democratizing” of machine learning, similar to what occurred with Big Data analytics. Machine learning is poised to become a massive business, with a market reporting stating growth of the industry is expected to move from $1.41 billion USD in 2017 to $8.81 billion by 2020, a CAGR of 44.1%.

 

Automation and easier to use platforms such as Amazon’s SageMaker and the Microsoft Azure Machine Learning Studio are offering machine learning tools to workers with little or no formal data scientist training. Even with automation, getting the most out of either platform requires human intervention to pick the right algorithms and craft models that are most likely to create predictive results. Firms looking at either the AWS or Azure platform should consider working with an IT consultancy that boasts experience with both choices, and can provide guidance on the right solutions to fit needs.

 

Comparing and Contrasting

 

Microsoft Azure and Amazon Web Services (AWS) are two of the core platforms for conducting machine learning on data held in the cloud. Amazon’s solutions is known as Amazon Machine Learning and uses algorithms to spot patterns found in a company’s data. These models are then used to generate predictions. The platform is highly scalable and can create billions of daily predications, and with no hardware or software investment required, firms can adopt a “pay as you grow.”

 

AWS also offers SageMaker, a fully managed service for data scientists who want to create machine learning models using their own data source and a choice of several learning algorithms. It also integrates with language frameworks including Apache MXNet and TensorFlow. AWS is attracting users to SageMaker with the trusted reputation of the infrastructure and the ability to leverage the entire full stack on AWS, all the way up to deployment. AWS’ additional benefits include no setup costs, speed of model creation due to automation, and the proven Amazon architecture. Drawbacks include limited prediction capacity and the amount of automation mean it can be difficult to use it as a machine learning trailing tool. Meaning, the automation does “too much” and leaves fewer tasks for a person trying to learn the underlying methodologies at work.

 

Each platform does have different data location requirements, with Amazon users required to have data stored in an AWS store before conducting machine learning modeling. With Azure, smaller data sets can be pulled from other sources (including AWS) but bigger data sets must reside in Azure.

 

Compared to Amazon’s platform, Azure is often seen as more flexible in terms of algorithms. This is a core benefit of Azure, the ability to support dozens of methods of data classification. Microsoft provides a “cheat sheet” to help data scientists to pick the right algorithms for the particular use case. So for example, the user might be guided towards supervised or unsupervised algorithms, logistic regressions, neural networks, or Poisson algorithms. The platform also offers the Cortana Intelligence Gallery that is a community-provided group of machine learning tools that are available to the broader Azure user community. The breadth of algorithms offered by Azure can make it a more appealing choice for experienced data scientists who perform more complex modeling. A drawback of Azure for machine learning is it’s not the best choice for speedily implemented projects, especially compared to AWS.

 

AWS and Azure both offer their own APIs that aid users in text analysis. For example, users can leverage Amazon Transcribe which is a tool for recognizing spoken text that can transcribe call center data or audio archives. Another tool is Amazon Polly which turns text into speech, which then allows companies to create unique voices for chatbots. Amazon Translate conducts translations using neural networks to convert multiple language into and out of English. Azure offers a similar group of APIs which it calls Cognitive Services which also offers speech and language tools for translate, speech to text, voice certification, and other capabilities.

 

 

The TechBlocks Machine Learning Advantage

 

TechBlocks can provide guidance to the technical staff that needs to validate and test cloud machine learning services. Making the best MLaaS choice for a business can be tricky, and requires a careful review of both short and long-term data analytics needs. TechBlocks’ experienced consultants understand the benefits of both Azure and AWS implementations, and can prepare tailored recommendations for every client. We’re a Gold Partner with Microsoft, a certified AWS integrator, and understand how to leverage both platforms for maximum gain. IT staff responsible for validating services on Azure or AWS should contact TechBlocks to discuss the best options for their cloud initiatives. Visit www.tblocks.com to learn more.