Trust Semantics Distillation for Collaborator Selection via Memory-Augmented Agentic AI

TL;DR

The paper tackles the overhead and latency of trust-based collaborator selection for resource-constrained devices in collaborative computing. It introduces Trust Semantics Distillation (TSD), a memory-augmented, LAM-enabled teacher–student architecture where a powerful server extracts task-specific trust semantics from global data and transfers concise knowledge to lightweight device agents. This enables rapid, task-aware collaborator selection with unified evaluation standards and reduced communication/computation on devices, using few-shot learning to adapt to dynamic environments. Experimental results show improved selection accuracy, reduced data collection, and faster decision-making, highlighting TSD's potential for scalable, reliable collaboration in heterogeneous networks.

Abstract

Offloading computational tasks from resource-constrained devices to resource-abundant peers constitutes a critical paradigm for collaborative computing. Within this context, accurate trust evaluation of potential collaborating devices is essential for the effective execution of complex computing tasks. This trust evaluation process involves collecting diverse trust-related information from every potential collaborator and performing trust inference based on the collected data. However, when each resource-constrained device independently assesses all potential collaborators, frequent data exchange and complex reasoning can incur significant overhead and further degrade the timeliness of trust evaluation. To overcome these challenges, we propose a task-specific trust semantics distillation (TSD) model based on a large AI model (LAM)-enabled teacher-student agent architecture. Specifically, the teacher agent is deployed on a server with powerful computational capabilities and an augmented memory module to perform multidimensional trust-related data collection, task-specific trust semantics extraction, and task-collaborator matching analysis. Upon receiving task-specific evaluation requests from device-side student agents, the teacher agent transfers the trust semantics of potential collaborators to the student agents, enabling rapid and accurate collaborator selection. Experimental results demonstrate that the proposed TSD model can reduce collaborator evaluation time, decrease device resource consumption, and improve the accuracy of collaborator selection.

Figures (6)

• Figure 1: A teacher-student architecture powered by LAM-enabled agentic AI for evaluating and transferring trust semantics between a trust server and task owners to enable task-specific collaborator selection.
• Figure 2: Workflow of the server-side memory-augmented teacher agent for evaluating and storing task-specific trust semantics. 1) The teacher agent collects devices’ resource information and stores it in the resource information storage component. 2) The teacher agent collects collaboration data and stores the collected data in the historical record storage component. 3) The teacher agent evaluates devices’ task-specific trust semantics based on historical collaboration records. 4) The teacher agent stores the evaluated task-specific trust semantics in the trust semantics storage component.
• Figure 3: Task-specific trust semantics transfer from the teacher agent to the student agent for collaborator selection. 1) The teacher agent retrieves the trust semantics of potential collaborators from the trust semantics storage component according to the task request initiated by the task owner. 2) The teacher agent retrieves the resource information of these potential collaborators from the resource information storage component. 3) The teacher agent performs task-resource matching analysis to assess resource trustworthiness and selects the trusted collaborators. 4) The student agent on the task owner selects the final collaborator using the trust semantics of trusted collaborators provided by the teacher agent.
• Figure 4: By utilizing the trust semantics stored in the memory module of the teacher agent, the proposed TSD achieves faster collaborator selection than baseline methods.
• Figure 5: The proposed TSD requires fewer data collection events than TMFSC, as its teacher agent leverages the augmented memory module to store collaborators’ historical performance records and trust semantics.