5 Scope
This Work Item will initially focus on the creation of a Technical Report analysing existing AI technologies (incl. Machine Learning) that can be resourced into oneM2M architecture. The TR will also investigate potential AI service use cases that use IoT data. The study on AI technologies and use cases will be further analysed to understand what features are supported and unsupported by the oneM2M system. Unsupported features will be used to generate potential requirements.
The Technical Report will investigate items as follows:
- State of the art AI technologies that uses data from IoT systems
- Potential use cases and requirements to support AI services and their data management
- Managing and manipulating training data in oneM2M to support AI technologies to build a model
- Managing and manipulating serving data in oneM2M to support AI technologies to run a model
- Feasibility study on running AI algorithms in oneM2M as a new CSF
- Generalization of steps performing AI algorithms to identify required common functions that can be supported by oneM2M
- Supporting different parameters schemes i.e., power consumption, cost for the future etc. in oneM2M to support AI services
- Distributed and federated ML on Edge/Fog oneM2M nodes
- Services to assist with deploying trained machine learning models onto field device nodes such that inferencing can be performed by the field device nodes using the models
This work item will also serve to capture input contributions generated from ETSI STF 601, started in Feb 2021. ETSI STF 601 objectives are to identify uses cases where IoT data and services require usability specifications. The data that IoT devices and platforms provide should be easily accessed, understood and acted upon by a large non-technical public in the case of humans (e.g. medical teams and their patients in the medical sector, mechanics in the automotive sector, first responders in the emergency sector, etc.) and by machines and processes when the data are fed to the AI components of a system (e.g. machine learning). This also means that the IoT technologies, devices and platforms themselves can be trustily used according to their initial objectives (e.g. easy installation, configuration, operation and maintenance). Based on these use cases, requirements and guidelines should be derived towards a horizontal cross-domain standard, with the specification of minimum requirements for usability of professional and general public IoT services, whether they are critical or not.
The work item will take into account of the activities in the following organizations:
- ETSI STF584 Artifical Intelligence for IoT Systems deliverables
- ETSI ISG SAI (Securing AI) regarding security of the AI/ML models
- 3GPP SA2 on AI/ML capabilities for SGC as Release 17 features
- AIStar R&D project in South Korea regarding data management and interoperability for AI/ML
Additional functionality will be identified through use case analysis and investigation of potential mechanisms.
Results of this WI are expected to propose changes for existing TS as CRs.