Shaohua Cao

Graph-based Multi-Agent Systems (MAS) enable complex cyclic workflows but suffer from inefficient static model allocation, where deploying strong models uniformly wastes computation on trivial sub-tasks. We propose CASTER (Context-Aware Strategy for Task Efficient Routing), a lightweight router for dynamic model selection in graph-based MAS. CASTER employs a Dual-Signal Router that combines semantic embeddings with structural meta-features to estimate task difficulty. During training, the router self-optimizes through a Cold Start to Iterative Evolution paradigm, learning from its own routing failures via on-policy negative feedback. Experiments using LLM-as-a-Judge evaluation across Software Engineering, Data Analysis, Scientific Discovery, and Cybersecurity demonstrate that CASTER reduces inference cost by up to 72.4% compared to strong-model baselines while matching their success rates, and consistently outperforms both heuristic routing and FrugalGPT across all domains.

As autonomous driving technologies continue to evolve, the demand for precise and prompt risk assessment in complex traffic scenarios has become increasingly critical to guarantee driving safety. Traditional driving risk field models typically trigger warnings only when vehicles enter each other’s risk zones, thereby failing to predict potential future spatial conflicts. This limitation significantly compromises the timeliness and reliability of warnings, particularly in multiagent scenarios such as intersections. To mitigate this limitation, we accordingly propose a dynamic driving risk assessment method (DDRAM) based on predictable multidimensional kinematic features. Based on the distributed digital twin information system architecture, a “query-centric” multimodal trajectory prediction algorithm driven by agent interaction is designed to achieve high-precision predictions of the upcoming motion states of traffic participants. By deconstructing the predicted kinematic features, we construct a multisource fusion model to dynamically assess collision risks. Furthermore, we design a quantifiable risk warning matrix to characterize potential risks with greater accuracy. This method significantly enhances the early warning capabilities in complex traffic scenarios. Experimental results demonstrate that DDRAM significantly increases the average precollision warning time (PCWT) by 76% and reduces the average precollision warning error (PCWE) by 72% compared with existing approaches.

With the rapid growth in demand for AI-generated content (AIGC), edge AIGC service providers (ASPs) have become indispensable. However, designing incentive mechanisms that motivate ASPs to deliver high-quality AIGC services remains a challenge, especially in the presence of information asymmetry. In this paper, we address bonus design between a teleoperator and an edge ASP when the teleoperator cannot observe the ASP’s private settings and chosen actions (diffusion steps). We formulate this as an online learning contract design problem and decompose it into two subproblems: ASP’s settings inference and contract derivation. To tackle the NP-hard setting-inference subproblem with unknown variable sizes, we introduce a large language model (LLM)-empowered framework that iteratively refines a naive seed solver using the LLM’s domain expertise. Upon obtaining the solution from the LLM-evolved solver, we directly address the contract derivation problem using convex optimization techniques and obtain a near-optimal contract. Simulation results on our Unity-based teleoperation platform show that our method boosts the teleoperator’s utility by $5 \sim 40\%$ compared to benchmarks, while preserving positive incentives for the ASP.

federated graph learning (FGL) has emerged as a promising paradigm for privacy-preserving collaborative learning in Social Internet of Things (SIoT), where nodes form complex interconnected networks. Existing FGL approaches face significant challenges including model degradation in handling nonindependent and identically distributed (non-IID) data and maintaining model performance across heterogeneous nodes. This article proposes framework via layer-wised clustering (FedLWC), a novel layer-wised clustering framework inspired by evolutionary processes is proposed to enhance the effectiveness of FGL. FedLWC designs three key aspects: 1) a fisher information matrix-based layer selection mechanism that identifies and evaluates critical model layers, which can reduce parameter redundancy; 2) a layer intersection clustering algorithm that preserves common key layers while accommodating local features; and 3) an adaptive layer merge strategy that effectively combines global shared layers with clustered key layers. To make sure that the proposed approach is rigorous, we conduct theoretical convergence analysis for the proposed framework under non-IID conditions. Extensive experiments on multiple benchmark graph datasets demonstrate FedLWC’s performance, achieving an average accuracy improvement of 7.01% compared to state-of-the-art federated learning methods.

Advanced AI-Generated Content (AIGC) technologies have injected new impetus into teleoperation, enhancing its security and efficiency. Edge AIGC networks have been introduced to meet the stringent low-latency requirements of teleoperation. However, the inherent uncertainty of AIGC service quality and the need to incentivize AIGC service providers (ASPs) make the design of a robust incentive mechanism essential. This design is particularly challenging due to uncertainty and information asymmetry, as teleoperators have limited knowledge of the remaining resource capacities of ASPs. To this end, we propose a distributionally robust optimization (DRO)-based contract theory to design robust reward schemes for AIGC task offloading. Notably, our work extends the contract theory by integrating DRO, addressing the fundamental challenge of contract design under uncertainty. In this paper, we employ contract theory to model information asymmetry while utilizing DRO to capture the uncertainty in AIGC service quality. Given the inherent complexity of the original DRO-based contract theory problem, we reformulate it into an equivalent, tractable bi-level optimization problem. To efficiently solve this problem, we develop a Block Coordinate Descent (BCD)-based algorithm to derive robust reward schemes. Simulation results on our unitybased teleoperation platform demonstrate that the proposed method improves teleoperator utility by 2.7% to 10.74% under varying degrees of AIGC service quality shifts and increases ASP utility by 60.02% compared to the SOTA method, i.e., Deep Reinforcement Learning (DRL)-based contract theory. The code and data are publicly available at https://github.com/Zijun0819/DROContract-Theory