TEUVONET

Simplifying AI on the edge

Evolution

Today, the core of IoT processing transpires in the Cloud. In contrast, Teuvonet employs simplified, patent-pending capabilities running on GPU or FPGA Systems on a Chip (SoCs) to train, test, and execute machine learning models near the IoT/IIoT Edge in real-time.


Architecture



  • Sensor Agnostic: Works with real-time, streaming (or batch) data generated by a one or more sensors of the same or different types. The main requirement for these sensors is that they have some means of transmitting the data.

  • Supports several Field Protocols: Handles sensor data transmissions via wireless (e.g. Wi-Fi, BLE, Zigbee, etc.) or wired protocols (e.g. Ethernet). Data can either be transmitted directly from a ‘smart’ sensor or indirectly by an edge gateway.

  • Machine Learning Capabilities: Supports key ML and AI algorithms including pattern classification, function approximation, clustering and segmentation, outlier and anomaly detection, and feature ranking and selection.
  • Learning and Inferencing at the Edge: All learning and inferencing takes place on commercially available FPGA or GPU-based SoCs. No need to transfer learning tasks to the cloud which improves latency and data transmission security. Once a model is learned, it is immediately available for execution on the same SoC, as well as available for incremental learning.

  • Results and Actions: Teuvonet generates reports and visualizations based on the results and insights derived from model training, testing and execution. These outputs can also be used to alert key operational personnel, to transmit actions to be executed by device actuators, or to transmit data and results to applications on the cloud.

Key Differentiators



Real - Time

FPGAs and GPUs enable real-time learning from streaming high-speed IoT/IIoT data

Secure

Because of its disconnected nature, localized learning close to the ‘Edge’ in communication-denied environments minimizes potential cyber attacks.

Explainable AI (XAI)

Produces human understandable IF-THEN rules that explain underlying reasoning and decisions.

Personalized/Individualized Models

Personalized models require less data and can account for the idiosyncrasies of a particular system.

Automated and Scalable Learning

Examines hundreds of models in parallel, finding the smallest feature set and simplest model that can best predict a phenomenon.

Machine Learning Capabilities



Pattern Classification

Determines to which class among a set of known classes a particular entity or process belongs based on a combination of sensor signals.

Outlier/Anomaly Detection

Determining the ‘onset’ of abnormal performance based on input from multiple sensors.

Function Approximation

Time series and regression modeling of data from multiple sensors used to predict or control future behavior.

Feature Ranking & Selection

Identifying which signals or variables are most important in producing observed values, patterns, segments, or anomalies.

Clustering, Segmentation

Divides entities or processes into separate groups based on similar values of sensor signals.

Applications

contact us

We would love to hear from you !

Please complete the form.
A Teuvonet representative will be in touch soon.


Feedback sent successfully!
Failed to send message! Retry!