Overview

The neural network framework does not create models with a predefined structure. It grows the network neuron by neuron, using a unique patented algorithm to adjust coefficients within a model. This approach enables users to build tiny and accurate models without any compression techniques. As a result, the models are:

• Well-optimized
• With fewer coefficients and neurons
• Up to 10 times smaller compared to other frameworks

You can deploy Neuton's models on the smallest MCUs, where they show fast inference time. In addition to Neuton, the Nordic Edge AI Lab supports building LiteRT models for the new Axon NPU, enabling you to leverage standard deep learning architectures. The platform also adds native Wake Word & Keyword Spotting support, enabling always-on, low-power voice-triggered applications for hands-free and context-aware device interaction.

The Nordic Edge AI Lab is easy to use and you do not need any special data science or embedded engineering experience. It automates the entire process of creating models for edge devices using innovative, cutting-edge technology. You can implement common TinyML tasks such as:

• Gesture recognition
• Wake word detection
• Keyword spotting
• Vital sign determination
• Human-machine interfaces
• Human activity recognition
• Machine fault classification
• Asset tracking and monitoring
• Any sensor-based time series task:
  • Regression
  • Classification
  • Anomaly detection