Thanks to a collaborative effort between Analog Devices and Antmicro, a new open-source platform is now making edge AI workflows more accessible and efficient for a broader range of developers.
Lowering the Barriers to Entry
At the heart of this innovation is AutoML for Embedded, a hardware-agnostic, open-source solution that automates the entire machine learning pipeline for edge devices. Whether you are an embedded engineer or a data scientist, this tool allows you to transform a dataset into a fully optimized AI model deployed to your chosen hardware, often in just minutes and without deep expertise in ML.
By sidestepping platform lock-in, the solution offers developers the freedom to deploy across any microcontroller unit (MCU), not just those from a single vendor. This flexibility is a major shift away from proprietary systems and opens doors for innovation across the embedded landscape.
Productivity-Boosting Features
- Visual Studio Code Plugin: Seamlessly integrates with the widely used VS Code environment.
- Kenning Framework Integration: Supports optimization, benchmarking, and deployment for a range of edge AI models and devices.
- Universal MCU Support: Optimized for Analog Devices’ Arm Cortex-M4-based MAX78002 and MAX32690 MCUs, but compatible with nearly any MCU.
- Automated Model Tuning: Utilizes SMAC and Hyperband with Successive Halving for intelligent search and tuning of model architectures and training parameters.
- Simulation and RTOS Compatibility: Features Renode-based simulation and Zephyr RTOS integration for both virtual and real-time development, minimizing the need for physical hardware in initial stages.
Streamlining the Machine Learning Pipeline
Deploying AI models to edge devices used to involve painstaking manual steps: data preprocessing, model selection, hyperparameter tuning, and extensive device-specific adjustments. AutoML for Embedded automates these processes, generating, optimizing, and evaluating multiple candidate models in rapid succession.
The platform checks each model against device memory constraints and benchmarks for size, speed, and accuracy. This makes it easy to select the most suitable model for deployment, even for those with limited machine learning experience.
Accelerating Open Source in Edge AI
Built on proven open-source tools, this platform encourages transparency, flexibility, and long-term control for developers. Renode and Zephyr RTOS integration further empower embedded engineers to develop, test, and deploy AI solutions in both simulated and real environments driving innovation and collaboration throughout the community.
Why This Matters
- Faster time-to-market: Transform datasets into deployable models in minutes.
- User-friendly automation: Minimal ML expertise required due to a streamlined, guided workflow.
- Vendor independence: Easily deploy models across diverse MCU hardware, avoiding lock-in.
- Insightful analytics: Detailed benchmarking reports aid in optimal deployment choices.
- Open-source roots: Extensible and community-driven for continuous improvement.
Takeaway: Broadening Access to Edge AI
AutoML for Embedded is a leap forward in making edge AI accessible and practical. By automating the most complex steps, it empowers a diverse audience, from embedded engineers to data scientists, to create, simulate, and deploy high-performing AI models on resource-limited devices. As open-source frameworks like Kenning, Renode, and Zephyr continue to mature, the vision of efficient, scalable, and accessible edge AI is coming to fruition.
Source: EE Times
Open-Source AutoML Tools Are Simplifying Edge AI Development