Differentiators

Cloud-Centric AI/ML

Edge-Centric
(Teuvonet AI/ML)

Real-time
  • Brains are in the Cloud
  • All model training and testing is in the Cloud
  • Model execution is usually in the Cloud but, depending on application, may be Edge based
  • Model training, testing, and execution are on the Edge
  • Eliminates the need to transfer data to and model from the Cloud
  • Enables incremental updates
Secure
  • Three major points of digital and physical attack:
  • The "Things" (devices, sensors, and actuators)
  • The Cloud to which the things are connected
  • The data being transmitted
  • Two major security concerns:
  • The "Things" (devices, sensors, and actuators)
  • The Edge device where the ML takes place
  • Data transmission to the Cloud is reduced
Explainable AI (XAI)
  • The focus is often on ‘deep learning’ where the large number of layers and features make it difficult to decipher ‘why’ particular results occurred and which features were most important.
  • Models result in IF-THEN rules, documenting the underlying reasoning and decisions, as well as determining which features were key
  • Elevates transparency and trust, guarding against compromised and malicious inputs
Personalized/ Individualized Models
  • Generally, results in ‘generic models tuned to an average target population.’
  • Generic models are not well suited for dynamic situations where models need to be ‘personalized’ to individual things (e.g. patients) and evolved overtime.
  • Handles the ‘generic’ case, as well as ‘personalized’ models that support incremental learning.
  • Personalized models require less data, learning from the current data of the system, monitor deviations from that state, and adjust accordingly.
Automated and Scalable Learning
  • ‘Deep learning’ is well-designed to handle complex problems, but they often require a ‘hugh amount of training data’, are extremely slow to solve, and involve substantial parameter tuning.
  • This makes them difficult to automate and usually requires both ML and domain expertise.
  • In Teuvonet the type of model has been (e.g. classification), the Teuvonet system automatically examines hundreds of models and features in parallel, finding the smallest feature set that best predicts the phenomenon of interest (i.e. autoML).