More organizations, entrepreneurs, and developers become interested in machine learning with each passing year as it opens many opportunities for business (forecasting, video and image recognition, deep text analysis, etc.). Some are just exploring possibilities, while others are already building fully-fledged ML-applications. Demand for simple and reliable solutions has been on the rise for a couple of years now and still continues to grow.
This means that service providers are interested in expanding and enhancing their products to satisfy the clients’ needs and cut the biggest slice of the market for themselves. In 2019, there are three leading players in the machine learning segment: Amazon, Microsoft, and Google. Each company has its own practices for machine learning. This segment has received a lot of attention over the course of the last decade, so we think it would be appropriate to take a closer look at the ML options provided by each corporation. We already did an overview of ML.NET for C# developers by Microsoft, and today we’re going to research the potential of Machine Learning services provided by Amazon.
A Bit of History and Facts
Before we start microscoping Amazon Machine Learning tools, we will give you a quick intro of Jeff Bezos’ brainchild. AWS was launched in 2006, which is a good headstart and a major advantage over the competitors. This allowed Amazon to meet the on-demand of enterprises concerning cloud computing services. Being first of your kind grants immense freedom to experiment and grow. That is why AWS is today is ahead of others with multiple locations worldwide. By the end of 2019, Amazon has 69 availability zones and 22 regions under the belt. The market share of AWS in 2019 has reached an astounding 33%, leaving Microsoft and Google far behind:
But you cannot always stay on top without progressing with your services. And when the star of Machine Learning started to shine prominently, Amazon couldn’t neglect the opportunity to arm itself with the new and very promising technology. As a result, the already massive range of Amazon Web Services has received an addition in the form of SageMaker, which was released in 2017. This tool was specifically designed to simplify the process of creating ML models and making them much easier to deploy. And that was just a beginning.
Since the very first implementation of ML services, Amazon proclaimed that their mission is to let every developer interested in the technology experiment with it freely. The company is constantly trying to simplify ML processes and make it accessible to a wider audience. Today AWS has the largest set of AI and ML services on the market. According to the company’s report, they have added 200 new features and services in 2019, giving truly unparalleled flexibility to developers and data scientists.
In addition to that, Amazon is improving the SageMaker tool that drastically accelerates the adoption of ML technologies. SageMaker is a managed service for building/training/testing/deploying ML models. It was created to help ML practitioners with those routines. The latest version of SageMaker shows astounding results with 10 times faster performance and inference costs reduction of 75%!
Amazon brings confidence that you’re using ML-optimized tools that will perform as you expect them to. Also, the organization provides all necessary complementing services to help you thrive in this department. Machine Learning requires not only ML tools but also security, data store, analytics, and compute services working together, and with AWS, you get a full set of comprehensive capabilities.
Amazon is the world’s #1 Machine Learning services provider. In 2019, the number of customers that use Machine Learning on AWS has exceeded 10.000, including many high-profile companies (Sony, Siemens, NVIDIA, Ryanair) and services (Tinder, Duolingo, Slack). Moreover, 85% of all TensorFlow cloud projects happen on AWS.
Amazon Machine Learning Services: What’s on the Menu?
AWS presents its ML capabilities as a three-layered stack. The segments that make this stack complete are:
- ML frameworks and infrastructure (TensorFlow, Amazon EC2, etc.).
- ML services (Amazon SageMaker and other tools to help with model training and deployment).
- AI services (Amazon Rekognition, Polly, Transcribe, and other solutions for adding AI to apps).
The frameworks layer contains deep machine learning algorithms and clusters that lay the foundation for ML projects. It aims at Machine Learning practitioners, specialists who are comfortable with designing their own tools/workflows to build and deploy ML models. Developers that already have some experience with machine learning will find a lot of fascinating options here. For example, optimized p3 and p3dn instances, which are the most powerful GPU instances available for machine learning on the open market.
The second layer of the AWL stack caters to ML-developers and data scientists. The main goal of these ML services is to simplify the Machine Learning workflow, make modelling easier with tools like the aforementioned SageMaker. The last or the top layer is a set of AI services that should allow you to be more effective in your everyday ML-related activities. These AI tools could be a perfect starting point for businesses that want to improve their customer experience (for example, make typical business operations more streamlined, less annoying, etc.). All services from the AWS ML stack are really useful, and there’s a huge number of features and options you can experiment with. Now let’s continue with the question that has probably formed in your head…
How to Start with AWS Machine Learning?
In order to make Machine Learning more accessible for developers, Amazon has created two unique services: DeepRacer and DeepLens. In addition to those two, the company also initiated their own Machine Learning training program. Now let’s talk about each option in-depth, so you will understand what options are available to you, how to access them, and how you can start your Machine Learning journey today.
But before everything else, we would recommend you to visit Amazon beginner’s guide on Machine Learning that is available on this page. It contains lots of useful information, tutorials, fact sheets, learning paths, explanation videos, and case studies, so it is totally worth your time. And now, let’s continue with the DeepRacer service.
It is a fully autonomous miniature of a race car designed by Amazon’s engineers to serve as an introduction to reinforcement learning. And it looks like this:
It was first introduced at the AWS re:Invent 2018 in Las Vegas, and today people all around the world are practising in deep races with it. The DeepRacer project is comprised of three main parts: the robocar, online racing simulator (powered by RoboMaker and SageMaker), and virtual racing league. Virtual and online racing segments are intended for people who don’t have the actual physical car model – you don’t need the car to participate in the DeepRacer League events. What makes this concept unique is its innovative approach to learn Machine Learning combined with reinforcement learning. The combination of learning and racing is a lot of fun and engages more people to try it out.
If you want to join the fun, you are going to need your own models. But don’t worry if you don’t have any because Amazon has thought of that too. There is a short training course you can get for free on the official website in the Racing Tips section. The guide will walk you through the basics, explain how to tune hyperparameters and provide various tips and tricks.
For developers eager to learn more about Machine Learning, Amazon invented DeepLens. In this article, we’re going to talk about the 2019 edition of DeepLens (the first one was released on November 29, 2018). It was the first device with AI that Amazon launched as a part of its Web Services. Basically, it is a wireless videocam, but with a twist – it is the world’s first wireless camera with deep learning capabilities. It lets a developer of any skill level to get started with deep learning. The best part is that you need less than half of an hour to run your first deep learning model (created with Amazon SageMaker or based on one of the pre-made samples).
The first version of DeepLens was only available in the USA, but with the release of the 2019 version, Amazon has added seven countries to the list. Five regions in Europe (the UK, Spain, France, Germany, Italy), one in Asia (Japan), and one more in NA (Canada). SageMaker has also received a significant update, which enabled it to run ML models two times faster.
With the 2019 update of DeepLens, it is safe to say that this device is a computer that lets you run computer vision models. It has 8GB of RAM, Intel Atom Processor with Gen9 graphics, and 16GB of storage space (Micro SD up to 32GB is available).
DeepLens comes with sample projects, so you can start building your own models straight out of the box. Users have access to models with options like object and activity detection, but nothing stops them from extending the functionality of the sample models. Machine Learning practitioners can go beyond and create their own models with the help of SageMaker.
As you see, it is a great platform for experimentation, advancing in the ML department, so it is not intended just for getting started with ML. For an easy start, however, you should check tutorials on the official website. For example, one of the newest tutorials allows you to build an application based on the Amazon Rekognition technology that will count the number of coffee cups workers drink at the office and build a leaderboard from that.
Amazon Machine Learning Training Program
This program gives every developer equal opportunities to learn Machine Learning and train with the same materials that Amazon engineers used while they were learning ML. The program is structured in such a way that you can access the specific workstreams according to your role: developers, data platform engineers, data scientists, and business owners. It includes thirty digital courses, and after passing the final exam, you can get an official certificate from Amazon. There are some extra courses that can prepare you for the test, so it is advisable to check them as well. Keep in mind that each ML speciality has its own progression path. Here is how Machine Learning path for developers looks like:
In the picture above, you can see different courses that Amazon recommends taking. In this particular case, everything starts with learning services and terminology and goes to more complicated topics such as Computer Vision Theory. No matter what speciality is, Amazon guarantees an exciting learning experience with lots of content. Thanks to this program, anyone can start making their first steps to the ML pro rank.
Amazon shows its dedication to extending Machine Learning services and make it accessible to every willing soul. Thanks to their tools and efforts the company has put into developing them, you don’t need to be a “rocket scientist” to start building ML apps today. Choose your learning path and make it to the end or switch between courses to check all available options. Whether you would like to start creating ML models now, whether you would rather learn in the hands-on mode with DeepRacer/DeepLens, or go through the educational content presented in training and certification modules. Amazon provides many fantastic options, and the choice is yours. For additional clarification of the topic or any other questions about machine learning, you can alwasy get in touch with our data scientists and developers from the SSA Data team.