Amazon Enhances SageMaker And Wants to be Your AI Provider
by Jim Lundy
Amazon re:Invent just ended in Las Vegas last week and besides the flurry of announcements around AWS Cloud and all the new Compute Services (I/O and Database), this year’s re:Invent was about Amazon wanting to be your AI provider. There were a number of enhancements to SageMaker, which is becoming its flagship AI offering for the enterprise. This blog provides a snapshot of what was announced.
Amazon Challenges Google and Microsoft
In what many will take as a surprise, Amazon is going hard after AI with its first Integrated Development Environment (or IDE) for the growing SageMaker product family, which is looking more like a platform. The Development Environment leverages a number of new tools that are targeted at developers. These include:
- Amazon SageMaker NoteBooks: this is Amazon productizing an OpenSource project called Project Jupyter. Amazon has been clever with this and many may not realize that Amazon is just leveraging existing Open Source AI initiatives, which it has done before.
- SageMaker Experiments is a tracking system for managing pilot and production AI deployments.
- SageMaker Debugger is for tracking and debugging issues with training Machine Learning Models.
SageMaker AutoPilot Is a Response to Google
With Google pushing automated ML model generation for the last two years, Amazon responded with its own SageMaker AutoPilot. AutoPilot allows for algorithm selection, tuning, data processing and I/O, which are handled automatically. Since AutoPilot was just announced, it is too early to compare it to Google’s Auto ML. However, Google AutoML does more when it comes to Computer Vision. We will also have to look at the ease of use of Amazon’s tools vs. Google.
Planning for A Large AI Bill
One of the things that many development and business teams have not done planning for is the charges they will face when using Cloud-based AI. For end users, this will be viewed as a cost of doing business. For vendors, the costs will significantly impact profitability. Aragon’s advice is to plan before you deploy. Don’t jump in headfirst into Cloud AI before you understand the true costs. If you think we are kidding, compare the cost of deploying standard TensorFlow Deep Learning, which is free or available as an enterprise offering from Google, versus the cost of SageMaker. This might be an eye-opening experience.
Bottom Line: Plan Before Going Live
Amazon is in it to win it in AI. The sheer processing demand required by AI algorithms makes SageMaker a future cash machine for Amazon. Enterprises need to understand costs before doing a production deployment of SageMaker or any of the other AI Cloud offerings.
Have a Comment on this?