By Daniel Pohl, Director of Product Management, Logicworks
Machine learning can be complex. However, it’s an increasingly critical part of business technology – and all technology professionals should make an effort to learn the basics.
In the last few years, AWS released AWS DeepRacer, DeepLens, and DeepComposer as fun tools to learn and teach Amazon Machine Learning. These products combine physical hardware with programmable machine learning, giving you real-world experience developing the AI brains behind self-driving cars, image recognition, and computer-assisted music orchestration.
At Logicworks, we’ve been experimenting with these tools over the last several years. Here are some fun results of our efforts.
What are AWS DeepRacer, DeepLens, and DeepComposer?
AWS DeepRacer is a 1/18 scale model racecar that drives itself. The DeepRacer AI engine uses reinforcement learning, a type of machine learning where the AI engine tries to earn “points” for accurately speeding around a racetrack.
My DeepRacer with custom skin, DeepLens, and DeepComposer
DeepLens is a camera with a built-in computer that uses “supervised” and “unsupervised” machine learning to recognize objects. With supervised learning, the developer first labels objects (e.g. “dog”) in training images so that DeepLens can then identify those objects. During unsupervised learning, DeepLens seeks to make sense out of patterns in images.
DeepComposer uses Generative Adversarial Networks (GANs) to autonomously create music. GANs are 2 machine learning models competing against each other – a “discriminator” model continuously evaluates the quality of music created by a “generator” model.
Again, the goal of these three services is not necessarily Amazon seeking to develop new innovative songs or cars – the point is to capture your imagination and educate participants in a hands-on way.
Start Your Engines…
I started using AWS DeepRacer at AWS Re:Invent 2018. I was excited to try this self-driving racecar that brings Python reinforcement learning programs (called “models”) to life. DeepRacer lets me compete either virtually or with real racecars on real racetracks at AWS Summits. My first Re:Invent model performed well on the virtual simulator but poorly on real racecars – the physical car using my model randomly veered off the track.
My son and I watching our DeepRacer compete at a New York City meetup
After Re:Invent 2018, I began teaching my 11 year old son Derek machine learning and Python coding – using DeepRacer as the learning platform. We benefited from a growing online community of DeepRacer enthusiasts sharing their experience and coding recommendations. By Summer 2019, Derek and I were ready to test our best reinforcement learning models with real DeepRacer cars at AWS Summit New York City. Our story – and our top 10 finish – are featured in this episode of DeepRacer TV. Our top 10 finish also earned us a new DeepRacer car!
My Logicworks colleague Mitch Fiorentini has achieved similar success in the DeepRacer virtual league. Mitch has achieved top 5% finishes in virtual races – quite a feat when the competition is the world’s top machine learning experts.
My son and I have also developed DeepLens image recognition models that identify animals and DeepComposer GANs that generate interesting (not quite good) music. These projects make use of the same Amazon SageMaker, Jupyter Notebook, and Python technologies that can be put to use for our Logicworks clients.
Bringing AI/ML to the Real World
All this learning (and fun) helps my Logicworks colleagues and I gain expertise in real-word machine learning architecture and development. Transportation companies use reinforcement learning to optimize their fleets, healthcare companies use image recognition to diagnose medical imagery, and advertising companies use GANs to predict human reactions. Logicworks has strategic clients in all these industries – and our understanding of these concepts helps us make our clients’ cloud environments smarter.
I strongly encourage you to check out these AI/ML learning tools for yourself. They’re fun and a great way to put Ai/ML concepts into practice! Ready to race? Look for me at the next New York City Machine Learning Community meetup or the next AWS Summit in New York City.
Logicworks is the AWS Premier Partner that helps companies build, manage, and optimize cloud workloads – whether they need advanced machine learning systems or auto-scaling fleets of virtual machines. Contact us.