This blog post was originally published at Edge Impulse’s website. It is reprinted here with the permission of Edge Impulse.
Today we are excited to announce the launch of Edge Impulse’s new auto machine learning tool, the EON Tuner! The EON Tuner helps you find and select the best embedded machine learning model for your application within the constraints of your target device. The EON Tuner analyzes your input data, potential signal processing blocks, and neural network architectures – and gives you an overview of possible model architectures that will fit your chosen device’s latency and memory requirements. Curious to observe the EON Tuner in action? In this or this public project you can see the results returned by the EON Tuner in our ‘Responding to your voice’ datasets.
Introducing the EON Tuner
While existing “AutoML” tools focus only on machine learning, the EON Tuner performs end-to-end optimizations, from the digital signal processing (DSP) algorithm to the machine learning model, helping you find the ideal tradeoff between these two blocks to achieve optimal performance on your target hardware.
The EON Tuner is an engineering tool designed to empower developers to select the best tradeoffs for their specific application. This task can usually only be achieved by a domain expert using complex model training systems and codebases.
Edge Impulse currently provides sensible parameter defaults for our DSP and neural network blocks, but there is no single “correct” model architecture for every use case. Real machine learning deployments have tradeoffs, and where you want to deploy a model matters. You might want to trade accuracy for less RAM or instead have firm latency requirements – each of these choices influences the best model.
Our primary design goal for the EON Tuner is to quickly help you discover machine learning architectures specifically tailored for your use case and dataset. The EON Tuner will also benefit your colleagues or anyone else you share your Edge Impulse project with who is not entirely familiar with the fine technical details of digital signal processing or neural network architectures. The Tuner eliminates the need for manual parameter selection to obtain optimal model accuracy, reducing user’s technical knowledge requirements and decreasing the total time to get from data collection to a trained model deployed on your edge device.
Future
Today you will find the EON Tuner tab located on your audio data projects within the Edge Impulse Studio. However, we’ll be adding support for other types of sensor data soon. With custom DSP blocks and the flexibility of our machine learning blocks (from classification to regression to anomaly detection), this also means that in the future you’ll be able to use the EON Tuner on novel sensor types and datasets.
The EON Tuner also gives the Edge Impulse engineering team a way to expose new signal processing blocks and machine learning algorithms to our developers. If we leverage a new keyword spotting algorithm from a promising research paper, we can add it to the Tuner and get instant feedback on its performance on real-life datasets (35,000+ projects on Edge Impulse right now!). At Edge Impulse, we constantly adapt to new advances in machine learning research, adding these advancements to the Studio and EON Tuner for your use. You can then explore/test the EON Tuner’s model advancements and easily update your primary model blocks with these cutting-edge enhancements and deploy to your target device.
Ready to get started with the EON Tuner? You can find a step-by-step guide here.
Jenny Plunkett
User Success Engineer, Edge Impulse
Mathijs Baaijens
ML Infrastructure Engineer, Edge Impulse
