Amazon simplifies machine-learning with AWS
(Image Credit: iStockPhoto/Pogonici)
Not content with letting Microsoft grab the headlines this week for cloud news, Amazon has debuted a new 'Machine Learning' service during the AWS Summit in San Francisco which intends to let developers build smart, data-driven applications which can "learn" based on past data.
Machine Learning is a complicated area of expertise which - unless you've spent years studying data science – is out-of-reach for most developers. Large companies such as Netflix have been using AWS for machine-learning about its users' viewing habits, but it's hard work and not everyone has access to the same level of resources and talent.
The same 92 percent accuracy as two developers did in 45 days, in just 20 minutes.
Amazon has set-out to fix this with their new service to make machine-learning accessible to any AWS-using developer. It first began as the sales-boosting recommendation engine found on Amazon's retail website which suggests other things you may want to buy based on other similar customers.
The firm realised that such predictions required expertise around things like statistical analysis, model building, data cleansing, and more. Even if you figure these out, you then have to work-out how to make predictions and implement them at scale…
Amazon has used their own experience in the field to reduce how machine-learning’s complexity and allow developers to quickly build and validate a model and process millions (or even billions) of predictions in real time. In fact, only three basic steps are required; 1) Build a model in Amazon S3 or Redshift, 2) Validate it, 3) Make batch predictions with trained models.
Large companies such as Netflix have been using AWS for machine-learning
In an impressive example, Amazon charged two developers with making their customer gender prediction better than their current 65 percent accuracy. After 45 days, the two developers managed to improve the accuracy to 92 percent.
This might sound impressive by itself, but Amazon then gave a single developer the same task but using Amazon Machine Learning. That same developer was able to achieve the same 92 percent accuracy as two developers did in 45 days, in just 20 minutes.
Amazon's machine learning service has some competition, however. Azure, for example, launched its own machine-learning tool back in February. IBM also boosted Watson's machine-learning capabilities last month with the purchase of AlchemyAPI.
Do you intend on trying Amazon's new machine-learning service? Let us know in the comments.