On-device learning, also known as on-device training, involves training or fine-tuning a machine learning model directly on the device where it is deployed, using real-time data from sensors. While on-device learning can be useful in specific scenarios, its utility is often limited due to the dependency on labeled data, which can be challenging (or impossible) to obtain in a deployed environment.

​
On-device learning with Edge Impulse

Edge Impulse provides several solutions for different on-device learning scenarios:
  • On-device learning for anomaly detection: Utilizes unsupervised learning to establish a “normal” baseline on-device, enabling the detection of anomalies in time series and visual data.
  • Zero-shot prompt modification with vision language models (VLMs): Allows developers to reconfigure vision models on-device by modifying prompts, facilitating real-time customization for vision applications.
  • Edge Impulse on-premise appliance: Enables customers to run the entire Edge Impulse platform on their hardware. This setup supports model training near production devices with human oversight for labeling.
These capabilities are available to enterprise customers upon request. For more information, please contact our sales team.