Ekkono Product Overview

We have developed an embedded software library – a Software Development Kit – built for the purpose to help developers rapidly and easily deploy edge machine learning, embedded onboard connected devices, to make them conscious, self-learning, and predictive.

The main functionality focuses on streaming analytics based on sensor data in combination with on-device learning. By having an integrated preprocessing pipeline, running a model on a device using one of the runtimes is extremely simple and only requires 5 to 15 lines of code, regardless of model or the amount of preprocessing.

The Uniqueness

What differs Ekkono’s technology from other machine learning techniques is the ability to do incremental learning on streaming sensor data – onboard the device. The benefits are significant:

  • No need to collect any data in advance.
  • Our software process high-frequency sensor data in real-time.
  • Data is processed onboard the device, which means product companies don’t have to send raw and potentially sensitive data, only relevant and enriched data, to the cloud for further analysis.
  • Radically reduced need for bandwidth and less dependency on the quality of the network connection.
  • Through incremental learning models adapt to local environments and learn their individual conditions.
  • Small memory footprint. You can even run Ekkono Crystal on a C64!

The Ekkono SDK

  • The software libraries are designed to efficiently provide edge machine learning and streaming analytics capabilities to platforms with constrained resources, with key features such as incremental learning and pipeline support. Seamless deployment from Python to C/C++.
  • Tools designed to guide through the somewhat complex task of selecting and evaluating different machine learning technologies for each individual use case.
  • Documentation such as Release notes, Implementation guide, API documentation and CRISP-DM Refined process documents.
  • Tutorials in Python on how to work with Ekkono Software for various tasks and use cases.
  • Reference implementations including example applications utilizing Edge libraries for various programming languages. 
  • Ekkono Studio – Online interactive environment based on JupyterLab containing tutorials and use case centric examples.

Built for purpose

  • Incremental learning for real-time continuous learning
  • Model-integrated data preprocessing pipeline
  • Conformal predictions
  • Automated change and anomaly detectors
  • Signal processing
  • Small memory footprint
  • No third-party dependencies
  • Hot swapping of models

Machine learning techniques used by Ekkono:

  • Linear regression
  • Regression trees
  • Random forest
  • Neural networks
  • Ensembles

Learn more about the Ekkono SDK