Decoupling Machine Intelligence from Application in IoT devices

Abstract

Currently, the most prominent model for developing intelligent applications for IoT devices is to have intelligence embedded into the application. This model is characterized by strong coupling between application logic and intelligence implementations in the code of the intelligent application. Alternatively, the intelligence can be taken out of the application and turned into a cloud service that application logic can utilize via standardized Web APIs. This model is characterized by weak coupling between application logic code and intelligence implementation. Strong coupling model makes lifecycle management of intelligence difficult. To update intelligence, usually the whole application must be updated. Cloud based weak coupling model also has multiple faults like the need for constant connectivity to the central cloud or data privacy concerns.

In this thesis, local on-device weak coupling model for building intelligent applications and its prototype implementation are presented. The model is based on the concept of intelligent layer. Intelligent layer is a layer between operating system and application layer that provides intelligent services to the processes in application layer. Presented prototype implementation is called intelligence layer service. It is able to serve limited type of machine learning models represented by Open Neural Network Exchange (ONNX) format.