TRUSTED DISTRIBUTED ARTIFICIAL INTELLIGENCE FOR CRITICAL AND AUTONOMOUS SYSTEMS

Intelligent systems are expected to adapt to change dynamically up to varying context by keeping trustworthiness of the system via available resources. In the case of decentralized resources and algorithms, scalability becomes a key issue compared to centralized approaches. The increasing number of nodes in swarm system impacts its behavior, since it utilizes complex computational mechanisms for co-operative missions, which has to dynamically fuse large-scale matrices of the nodes and merge those to be able to scale and train the system in (near) real-time. The literature review in the domain shows that keeping the system resilient is challenging since it requires real-time updates and predictions in different system layers. In this context, some approaches based on ledger chain structures and big-data technologies can accomplish the transaction scalability and memory-speed analytic performance in some manner. Despite that, mission/safety/operation-critical applications (such as cooperative autonomous missions), require measuring and monitoring trust in critical checkpoints of a system in operation while keeping the expected performances of the total system. The thesis aims at solving such challenge and targeted the development of a new Methods called Trusted Distributed Artificial Intelligence (TDAI). The main contributions of the thesis are on the development of novel method that covers the following aspects:
- Measuring, quantifying and justifying trust in distributed systems.
- Enabling trusted scalability of autonomous systems.
- Assuring trust for swarm intelligence mechanisms.
- Manipulating swarm system units with search and mining focus to implement TDAI methodology in
(near) real time.
In order to develop such a holistic methodology approach, we gradually contributed from different perspectives. We initially proposed to utilize a holistic MEMCA (Memory-Centric-Analytics) abstraction, to maximize the trust factor of the system while enabling trusted scalability of the transactions and memory- speed. Data locality is extended to the edges in trusted scalable manner to handle the complexity of data/transaction-flow while the memory performance was ensured in the swarm. This approach is based on micro-services architecture and has innovative approaches for layer-wise structure enabling verification of trust in critical checkpoints of an operational system via observed and observing nodes that can enable us to build growing intelligent mechanisms with software-defined networking (SDN) features by virtualizing network functionalities with maximized trust features for continuous trust monitoring in observed context.
Accordingly, critical feature sets of AI systems and growing intelligent mechanisms are identified, measured, quantified and justified dynamically with defined three architectural perspectives (1) central, (2) decentral/autonomous/embedded, (3) distributed/hybrid for emerging trusted distributed AI mechanisms. Therefore, resiliency and robustness can be assured in a dynamic context with an end-to-end Trusted Execution Environment (TEE) for growing intelligent mechanisms and systems. Thanks to TDAI methodology, the system could consider the trust indicators that can bring a confidence on the predictions in the distributed context. This way, distributed algorithms can be processed at massive scale by ensuring trusted scalability. Therefore, resiliency and robustness can be assured in a dynamic context with an end-to-end Trusted Execution Environment (TEE) for growing intelligent mechanisms and systems. Besides that, the trust measurement, quantification, and justification methodologies on top of TDAI are also applied in emerging distributed systems and their underlying diverse application domains. Finally, smartness features are also improved with human-like intelligence abilities at massive scale thanks to the promising performance of TDAI at massive scale initial deployment experiments. TDAI is demonstrated in a cross-border financial risk monitoring scenario within the distributed systems formed by the connected nodes, to detect and minimize critical risk alerts within the observed context. The main objective is to maximize trust values of the critical nodes and the observed environment within the monitored time-span.
Following the thesis, the simulation of autonomous and connected vehicles will be pursued as the application domain with specific use-cases for further massive scale deployments within different cross-border challenges. Furthermore, the innovative approaches will also be utilized to wider spectrum intelligent system requirements of smart-cities and space systems as well with improved trust features.