Dynamic Trust Provisioning for Collaborative Task Completion and Value Realization in Future Connected Systems
October 22 @ 6:30 pm - 7:30 pm
The rapid evolution of communication and computing technologies and their integration with Artificial Intelligence (AI) have enabled a plethora of emerging vertical applications, leading to increasingly complex tasks. Effectively completing such tasks for application-oriented value realization requires fulfilling diverse task-specific needs, which in turn rely on the tailored provisioning of heterogeneous services through efficient resource allocation. However, the limited onboard resources and service capabilities of individual machines make it impractical for them to complete tasks independently. As a result, these considerations necessitate seamless collaboration among interconnected machines for application-oriented value realization by dynamically matching the comprehensive task completion requirements with resources and capabilities of potential collaborators through proper collaborator <a href="http://selection.
To” target=”_blank” title=”selection.
To”>selection.
To address the related challenges, this thesis first introduces a new concept of task-specific trust as a holistic evaluation metric and task completion mechanism, guiding collaborator selection process to ensure their resources and capabilities are precisely aligned with diverse task-specific requirements, thereby enabling needs fulfillment and ultimately realizing application-oriented value. To realize this objective, dynamic trust provisioning approaches are proposed to select collaborators for effective task completion by capturing task-specific needs, profiling the task owner’s resources and capabilities, and evaluating their external collaboration <a href="http://conditions.
In” target=”_blank” title=”conditions.
In”>conditions.
In pursuit of value realization in time-sensitive tasks from Internet-of-Vehicles (IoV), we propose a Rapid Trust Evaluation (RTE) mechanism to enable rapid and accurate collaborator selection for fulfilling the timeliness need of task completion. RTE accelerates collaborator selection in cold-start situations by leveraging indirect trust, generated from trusted recommendations of network servers or reliable vehicles. To improve accuracy without sacrificing timeliness, an adaptive aggregation scheme progressively incorporates task owners’ collaboration experiences and environment observations as direct experiential and capability trust, with all trust factors related weights dynamically adjusted across different IoV collaboration <a href="http://stages.
Beyond” target=”_blank” title=”stages.
Beyond”>stages.
Beyond time-sensitive tasks, the proposed task-specific trust is further applied to completing complex tasks with diverse requirements, where value is realized through selecting collaborators that fulfill multi-dimensional task-specific needs. These needs are modeled as distinct metrics, termed Value of Task Completion, through which task-specific trust guides collaborator selection. A trust-guided bipartite graph matching problem between tasks and collaborators is then formulated, where matching complexity is reduced by initially clustering tasks into limited categories and subsequently arranging them by <a href="http://priorities.
To” target=”_blank” title=”priorities.
To”>priorities.
To further reduce risks for task completion in highly dynamic systems, task-specific trust is integrated with Graph Neural Networks (GNNs) as a new trust prediction method that effectively leverages historical information. However, intensified resource contention under limited resources and concurrent tasks limits its effectiveness in fulfilling task-specific needs and thereby undermines value realization. To address this, a trusted collaborator group formation and task allocation strategy is proposed for concurrent task completion under long-term fairness. The concurrent task orchestration is formulated as a long-term average task utility maximization problem under α-fairness, solved via a hierarchically matching strategy that is supported by task-specific trust prediction and fairness-aware Deep Reinforcement Learning (DRL), to form and match trusted collaborator groups in a coarse-to-fine <a href="http://manner.
Virtual:” target=”_blank” title=”manner.
Virtual:”>manner.