One of the best features in the Edge Impulse platform just got better; The EON™ Tuner has been updated to use Bayesian optimization (previously random search). This is great news, as this delivers better impulses in less time while using less compute!
The EON Tuner is a powerful engineering tool that helps you find and select the best performing impulses for your application, given your hardware target resource constraints like RAM, flash, and inference time. This saves you the time and effort of manually creating experiments and allows you to explore a parameter search space more efficiently.
One aspect of the EON Tuner that sets it apart from other tuners is that it includes parameter selection for your input and processing blocks in addition to model hyperparameters for your learning block; it is not simply another model hyperparameter tuner. This allows you to optimize your feature extraction in conjunction with your model, ensuring the two are closely paired.
Bayesian optimization: An intelligent algorithm
There are several approaches to exploring a parameter search space. For example, the most common methods include: manual search, random search, grid search, and Bayesian optimization. Compared to the other techniques listed, Bayesian optimization is a more efficient and effective option. It is a "smarter" algorithm that learns from prior results by building a probabilistic model based on these results, the possible parameter values, and your objectives to determine what area of the search space to explore next. This leads to improved results in a shorter amount time using fewer compute resources.
To learn more, please visit the EON Tuner documentation.
Get started with Edge Impulse
If you don’t have an Edge Impulse account already, you can sign up for a free account at edgeimpulse.com/signup. It only takes a minute. Once you have an account, two great resources to get started with are our documentation and forum.
Please feel free to also join our Edge Impulse Developers Discord sever to connect with a community of innovators, enthusiasts, and industry experts building the future of edge AI together. We can’t wait to see what you create!