The key is Antmicro’s Visual Studio Code plug-in, which is built on its Kenning machine learning library.
“Kenning’s aim is not to bring yet another training or compilation framework for deep learning models,” according to Antmicro. Kenning instead provides “a unified API that focuses on deployment tasks rather than their implementation – the developer decides which implementation should be used for each task. This way, switching to another target platform results, in most cases, in a very small change in the code, instead of re-implementing larger parts of a project.”
ADI is calling its version ‘AutoML for Embedded’, and has integrated its interface into Analog’s CodeFusion Studio IDE (integrated development environment).
The Cortex-M4 microcontrollers supported are MAX78002 and MAX32690.
“Enabling workflows based on proven open-source solutions is the backbone of our end-to-end development services,” said Antmicro v-p of business development Michael Gielda. “With simulation using Renode and integration with Zepher RTOS, the road to edge AI development using AutoML in Kenning is open.”
The tool automatically searches for models that fit the supplied dataset, using ‘SMAC’ (sequential model-based algorithm configuration), Hyperband and ‘successive halving’.
Model size is verified against the particular MCU’s ram, and then candidate models can be optimised, evaluated and benchmarked using Kenning’s standard flows.
Reports on size, speed and accuracy are available to guide deployment decisions.
“In a recent demonstration, the tool was used to create an anomaly detection model for sensory time series data on the MAX32690. The model was deployed both on physical hardware and its digital twin in Renode simulation,” said ADI. “Other potential applications include: image classification and object detection on low-power cameras, predictive maintenance, natural language processing for text analysis, and real-time action recognition in sports and robotics.
AutoML for Embedded is available now on Visual Studio Code Marketplace and GitHub
