QoS-Aware Token Scheduling and Private Data Valuation for Multi-Modal Agentic Networks

The paper introduces a novel approach for fair token allocation and private data valuation in decentralized agentic systems, addressing the challenges posed by data centralization and heterogeneity. It proposes embedding multi-modal representations in a shared semantic space and utilizing differentially private prototypes to maintain utility while minimizing semantic leakage. Key findings include improved contribution-based fairness and quality of service (QoS) in simulations, along with enhanced privacy protections against image reconstruction attacks, which are crucial for practitioners focusing on data governance and privacy in AI applications.