How Wallaroo Makes Edge Machine Learning More Accessible
November 15, 2021
Explore the power of edge machine learning with Wallaroo’s advanced SDK, making AI deployment leaner and more accessible. Dive into real-world applications and discover how Wallaroo’s unique approach offers faster, secure, and cost-efficient solutions for even the most challenging edge environments.
Are you looking to tighten security, lower costs, and make faster data-driven decisions? Then you might want to consider edge machine learning (edge ML).
In brief, edge ML is machine learning applied on or close to the device that captures the data. Cloud ML typically comes with large cloud computing costs, whereas edge ML can save you both bandwidth and expenses. Learn more about the benefits of challenges of edge machine learning.
Your weakest point in any edge ML setup will typically be your data endpoints. Whether you’re working with cobots in a factory line, security cameras, or fighter jets, most resources in each unit are used for purposes other than machine learning. This led us to build a lean deployment model for edge ML that makes enterprise-grade production AI faster and widely accessible for every organization
Wallaroo SDK – simple and fast production AI
We teamed up Wallaroo’s signature lightning-speed ML processing with the robust and highly available Kubernetes, so you can bring advanced edge ML to any device — even those with minimal RAM. We put all you need into the Wallaroo SDK and API.
With our lightweight Kubernetes-stack, you can set up ML standalone or with edge connectivity. You can also run standalone machine learning in environments that seldom have an internet connection. These can be environments deep underground, in international waters, or with such high security that they can’t be directly connected to the internet. Some of the use cases we’ve seen are:
- Military vessels like submarines and fighter jets generate petabytes of IoT data every day with limited internet access – for practical reasons as well as security reasons
- Inference on data collected around or on cobots in a high-security manufacturing setting
- Security scenarios for autonomous vehicles that need to be able to run machine learning and either keep going despite gaps in internet connection or take action in traffic to avoid dangerous situations
The resulting data can be uploaded for central logging and learning when the edge deployment is complete. The lightweight Kubernetes containers are faster to transfer and therefore have a lower risk of exposure.
You can use the Wallaroo API and SDK for all your edge deployments. Even if you’re building a complex pipeline, running A/B tests, or doing other experiments in your edge environments. One of our customer’s favorite things about using our solution for edge ML is that they can easily reproduce entire pipelines and audit logging.
Deploy, scale, and distribute in one place
If you’re planning to deploy edge ML in an environment that’s mostly connected to the cloud, your centralized Wallaroo cluster will handle most of the work. The cluster technology is also lightweight – usually running KubeEdge – to keep your membership costs low.
You can deploy, scale, and distribute models as well as manage a complete log from one centralized platform.
Plus, since the Wallaroo Edge stack is built with minimal dependencies and footprint, you can finally stop worrying about the usual bottlenecks in your system like limited capacity in your data endpoints or limited bandwidth for data transfers.
Wallaroo also runs where no cloud connection can go
More environments can’t be connected to the internet than we listed above. Whether this is for practical reasons or security reasons, Wallaroo has you covered. You can deploy our edge ML solution separately – also known as air-gapped. Simply load software on a USB stick or other separate drive and send a technician to set up the Wallaroo Edge node on-site.
Run your edge nodes managed or unmanaged to fit your scenario
In this case, the Wallaroo stack will be compiled to a single binary providing the full Wallaroo API and SDK. From this version, you can compile our popular Kubernetes-stack to get a managed Edge node up and running.
In an unmanaged edge use case, you do most of the work. All automation and managing the models, pipelines, and logs must be done separately. You can follow the file system directories to easily deploy the models and pipelines.
Are you wondering how our solution will work for your edge ML use case? Reach out to us for a demo and answers to all your questions.
Reach out to us at deployML@wallaroo.ai to learn more.
