Artificial intelligence

AI3 focuses innovative solutions in the AI domain, addressing aspects that range from real-time interactive learning to training and usage of AI systems across organisation domains. The goal is to enable AI systems to learn continuously based on the experiences of other components in the same domain in the cases of emerging situations / context. The latter will shift learning from a sporadic / offline process to a continuous real-time process enabling AI systems to adapt to asynchronous changes. Furthermore, AI3 provides capabilities considering datasets owned and managed by different organizations in order to enable training and exploitation of outcomes without compromising confidentiality and privacy properties. Towards AI outcomes of high quality, AI3 researches on techniques to to extract latent features that correlate different fields, and identify possible defects in the datasets. By monitoring the frequency of the errors from a specific source, possible deterioration and levels of data veracity are identified and correlated with the information regarding the obtained datasets and its corresponding quality properties in order to deal in a holistic way with potential noisy inputs.

Data management and analytics

AI3 provides approaches for efficient real-time big data analytics through workload optimized data stores and query parallelization techniques that can also be applied to complex events (i.e. CEP). The aforementioned approaches enable data analytics to be performed both on streaming and on stored dataset through a powerful environment for data management on top of different data stores (i.e. SQL, NoSQL, CEP). Furthermore, AI3 provides technologies for incremental queries, thus enabling partial results to be obtained in advance. Data analysis functions are applied to specific snapshots of the data to obtain partial results, while continuous updates on the data trigger the incremental data analysis algorithms.

Cloud and edge computing

AI3 focuses on novel architectures for the efficient development and placement of distributed edge applications over combined IoT and cloud computing ecosystems. Service mesh architectures are integrated with cloud and edge computing architectures in order to allow optimized and resource-decoupled utilization of the underlying cloud and edge resources. To this end, a complete AI3 solution addresses in an end-to-end way the requirements of distributed edge applications: from the abstraction of the underlying infrastructure, through the collection and real-time processing of the data, to the development of the application to exploit the edge data while managing the provisioned levels of users experience.

5G networks and applications

A wide set of key 5G technologies and enablers are targeted by AI3. These include adaptable orchestration of VNFs considering 5G applications contextual information, slice intents and emerging resource parameters in real-time. What is more, AI3 enables the modelling of 5G applications and the capturing of their requirements in a structured way (through a set of metamodels) in order to ensure optimum utilization of 5G underlying infrastructures. In the context of 5G, AI3 also delivers a solution for the verification and validation of 5G applications (in terms of composite graphs that consist of VNFs) in terms of functional and non-functional requirements, for a set of KPIs relevant both to the application and to the underlying 5G infrastructure elements.

Application analysis and workload prediction

AI3 has a long record relevant to the analysis of application and the provision of performance estimates for different workloads. A complete application analysis and dimensioning framework allows the analysis and estimation of resource requirements for non-monolithic applications, by taking into account the interactions between the application components / micro-services, the data flows, the workloads per case and per application component, the network overheads and the cascading effects of these parameters on the application performance. Resource estimates are modelled and incorporated in elasticity models to facilitate real-time adaptation of resource provisioning according to the extracted application analysis and dimensioning knowledge.