At its 2018 re:Invent Conference, Amazon Web Services (AWS) threw down the gauntlet by launching the AWS DeepRacer, a 1/18th scale race car. Using reinforcement learning, the AWS DeepRacer allows developers—and, in our case, incredibly bright students from the College of Charleston—to get up close and personal with artificial intelligence.
Reinforcement learning, according to AWS, has a super power: “it learns very complex behaviors without requiring any labeled training data and can make short-term decisions while optimizing for a longer-term goal.” We wanted to help mentor students by challenging them to out-race the racer using machine learning.
Making the AI grade: Putting the AWS DeepRacer to the test
We teamed up with College of Charleston data science majors, Joshua Turner and Eliza Starr, to demonstrate Logicalis’ ability to provide services and machine learning inference at the edge.
The AWS DeepRacer car uses reinforcement learning to teach it to stay on the track. It takes a picture of its position on the track and, through reinforcement learning, adjusts its direction and speed based off its performance.
After improving the model’s parameters, the students trained our virtual car 42 (for the uninitiated, 42 is the Answer to the Ultimate Question of Life, The Universe, and Everything from The Hitchhiker’s Guide to the Galaxy)on each track. The result? 42 outperformed the AWS DeepRacer in finishing time, number of laps completed, and cumulative reward.
Artificial intelligence: Use cases and real-world applications
In addition to showing Eliza and Joshua’s creativity, the project demonstrates our ability to apply machine learning at the edge in applications such as predictive maintenance, autonomous driving, image and sound recognition, surveillance, and anomaly detection.
Based on a recent survey, IDC listed the top use cases for IoT project-based services as:
- Freight monitoring
- Security systems
- Smart home
- Inventory management
- Wearables (health, banking)
- Remote health monitoring
- Manufacturing operations
- Customer insights (communication, demand analysis)
- Connected vehicles
- Smart building
These real-world examples show the value of machine and deep learning in transformation efforts:
- Rolls-Royce feeds data collected from more than 70 trillion data points in its engines into machine-learning systems that predict when maintenance is required.
- Elevator manufacturer thyssenkrupp feeds data collected from its 1.1 million elevators worldwide into trained machine-learning models, allowing thyssenkrupp to receive real-time updates on elevator status and predictions on which are likely to fail.
- John Deere, the agricultural equipment manufacturer, is using deep learning techniques to design equipment that can make real-time adjustments in the field to maximize crop yields.
Clearly, data used well can deliver transformative benefits. Yet, many organizations are either not collecting data or collecting massive amounts of data that are going unused—data that can help drive business transformation objectives such as lower costs, increased efficiency, and improved productivity. What are your business transformation goals—and what is your organization doing to achieve them?
Logicalis: Using data—and machine learning—for good
With a dedicated IoT & Analytics practice staffed by highly skilled and certified engineers and experts, Logicalis has made critical investments in key partnerships with major cloud platform providers—including AWS, Microsoft and IBM—as well as specialty partners that play in the IoT sandbox. And Logicalis recently achieved Cisco IoT Authorization, among the first of Cisco’s partners to do so.
We believe in using machine learning and deep learning to help solve big problems in healthcare, manufacturing, government and education. And we’re actively working with clients to improve patient outcomes, increase plant efficiency and safety, increase government efficiency, and improve student engagement.
Join our movement. Contact us to get started on solving your biggest issues. Help use #dataforgood.