Organizations: School of Automation, Guangdong University of Technology, Guangzhou 510006, China · Department of Computer Science, City University of Hong Kong, Hong Kong · School of Information and Communication, Guilin University of Electronic Technology, Guilin 541004, China · Anhui Engineering Research Center for Agricultural Product Quality Safety Digital Intelligence, Fuyang Normal University, Fuyang 236037, China · School of Physics and Electronic Engineering, Fuyang Normal University, Fuyang 236037, China · Department of Electrical and Computer Engineering, University of Manitoba, Winnipeg, MB R3T 5V6, Canada
Artificial Intelligence-Generated Content (AIGC) services employ Generative AI (GenAI) models to automatically generate diverse content. Mobile AIGC networks host GenAI models on edge-located AIGC Service Providers (ASPs) to deliver low-latency and personalized AIGC services for mobile users. However, AIGC inference tasks typically occupy GPU memory until task completion, causing GPU memory exhaustion at serving ASPs and triggering out-of-memory failures rather than merely increasing service latency. Existing studies on AIGC task orchestration have largely overlooked GPU memory feasibility constraints. To address this issue, we develop a cooperative multi-agent orchestration framework, in which each edge node is equipped with a scheduling agent to route tasks to local ASPs or neighboring edge nodes. Since scheduling agents make decisions based only on local observations, while peer offloading couples their resource states and long-term utilities, we formulate the orchestration process as a cooperative Decentralized Partially Observable Markov Decision Process (Dec-POMDP). To solve the Dec-POMDP, we propose an \underline{A}ttention-guided \underline{M}ulti-\underline{A}gent deep reinforcement learning algorithm with \underline{Diff}usion policies (A-MADiff) under the centralized training with a decentralized execution paradigm. A-MADiff employs diffusion-based decentralized actors to generate multi-modal preferences over feasible orchestration actions, and an attention-guided centralized critic to estimate per-agent values from cross-agent states under GPU memory heterogeneity. Numerical results demonstrate that A-MADiff significantly improves the cumulative reward over the state-of-the-art baseline.
Figures & tables
Fig. 1: Overview of the proposed GPU memory-aware task orchestration framework in distributed mobile AIGC networks.
Fig. 2: Overall architecture of the proposed A-MADiff algorithm for GPU memory-aware AIGC task orchestration.
Category
Parameters
Values
System
Scheduling agents N
20
ASPs per agent ∣Kn∣
5
Queue capacity Qmax
50
Max offloading hops Hmax
2
Task
Arrival rate λn (per step)
0.3
Pre-generated task pool per agent
500
TABLE I: Simulation parameters.
Methods
Cumulative Rewards
Effective Success Rates
Rewards per Completed Task
Negative Completion Rates
OOM Rates
Heuristic Methods
Random
−407.78±19.96
8.18%
−0.0577
42.46%
5.78%
RoundRobin
−373.35±19.41
7.94%
−0.0923
45.97%
4.85%
Greedy
281.84±19.13
77.71%
0.2414
22.29%
2.07%
MADRL Methods
MADDPG
355.81±19.32
87.31%
0.3260
12.68%
1.56%
MAPPO
422.38±31.35
90.87%
0.4036
8.94%
1.04%
MASAC
476.58±16.59
91.29%
0.4278
8.68%
1.35%
TABLE II: Performance comparisons of A-MADiff and baselines. Bold values indicate the best result in each column.
Fig. 3: Training performance of all evaluated methods. The heuristic baselines provide constant reference levels, the MADRL methods are evaluated at regular intervals on separate evaluation episodes during training, and the vertical axis in (c) uses a logarithmic scale.
Fig. 4: Action composition of different MADRL methods during evaluation, showing the ratios of local execution, peer offloading, and queue deferral.
Fig. 5: Reverse denoising process of the diffusion-based actors within one joint scheduling decision at decision step t=58 . Arrow width and opacity encode the probabilities obtained by applying the masked softmax in ( 29 ) to the intermediate logit vector xn,ℓt at denoising step ℓ∈{20,16,12,8,4,0} , and only the two largest offloading candidates of each agent are drawn. Node color encodes the available GPU memory ratio in ( 10b ). Blue, purple, and orange arrows denote the preferences of scheduling agents n2 , n8 , and n3 , respectively, and solid arrows in the final panel mark the selected actions.
Fig. 6: Mean attention weights of the centralized critic at decision step t=7 , averaged over the four attention heads and the two critics. Rows and columns correspond to the query agent n and the key agent n′ in ( 31 ). Green boxes mark the three largest weights in the row of n2 , and the orange dashed box marks the offloading target selected by the actor.
Fig. 7: Robustness comparison of the cumulative reward under three scaling dimensions. The GPU memory capacity scale uniformly scales the GPU memory capacity of all ASPs, the GPU memory requirement scale uniformly scales the GPU memory requirements of all tasks, and the task arrival scale uniformly scales the task arrival rate at all scheduling agents. Each experiment varies one dimension while the others remain at their default values.
Large language models (LLMs) enable autonomous agents for reasoning, planning, and tool use. Recent systems increasingly organize these agents as graphs of specialized, interconnected nodes. Although graph-based orchestration supports flexible decomposition and coordination, it creates a key challenge: \textbf{attention allocation}. As workflows grow, existing approaches often execute graph components uniformly, wasting resources on irrelevant or low-impact tasks. We introduce \textbf{Attention Orchestration}, a paradigm that extends Transformer-style attention from token representations to workflow-level agent coordination. Our framework, \textbf{Adaptive Goal-aware Attention Orchestration (AGAO)}, dynamically estimates agent importance based on user objectives, graph dependencies, and computational constraints. AGAO combines three components: (1) goal-aware attention, measuring semantic relevance between user goals and agent capabilities; (2) topology-aware attention, modeling structural dependencies in agent graphs; and (3) resource-aware attention, allocating budgets and execution priorities across heterogeneous agents. Together, these mechanisms transform static agent graphs into adaptive systems that focus computation on goal-critical reasoning paths. Experiments across diverse multi-agent workloads show that AGAO improves task effectiveness while reducing unnecessary computation, latency, and token consumption compared with existing graph-based execution strategies. Our work establishes \textbf{Attention Engineering} as a direction for scalable, intelligent multi-agent systems. Code: https://github.com/MingzhouFan97/AGAO.
Mixture-of-Agents (MoA) systems improve reasoning accuracy by routing each query to multiple expert LLMs and aggregating their outputs. Efficiently executing this workload on limited GPU resources has bottlenecks. Skill-based routing creates skewed expert demand, and combining instruction-tuned LLMs with long-reasoning models results in extreme variability in generation lengths. Consequently, traditional scheduling strategies suffer from significant GPU idling and throughput collapse due to load imbalances. We present MOSAIC, a scheduling framework to accelerate MoA workloads. First, we formulate an Integer Linear Program (ILP) based scheduler that jointly optimizes expert placement and per-worker prompt assignment from offline-profiled costs, replicating reasoning experts across workers while pinning lightweight ones. Second, MOSAIC uses confidence-aware adaptive aggregation, leveraging inter-expert agreement to bypass the heavy final aggregator LLM for consensus queries. In our 4-GPU system, MOSAIC achieves up to 2.5x expert-stage, 4.23x aggregator-stage and 1.7~2.3x end-to-end speedups over the baseline scheduler, while matching accuracy within 0.1pp.
Saptarshi Mitra, Yifan Zhang, Rachid Karami +5
University of California, Irvine, USA · University of Southern California, Los Angeles, USA · Intel, USA
Artificial intelligence-generated content (AIGC) has emerged as a transformative paradigm for automating the creation of diverse and customized content, giving rise to rapidly growing computational workloads in cloud data centers. It is imperative for AIGC service providers (ASPs) to strategically schedule AIGC workloads to reduce data center energy costs while guaranteeing high-quality content generation. However, the distinctive characteristics of AIGC services pose critical challenges, including model heterogeneity across ASPs, implicit service quality evaluation, and complex inference process control. To tackle these challenges, we propose a joint energy management and coordinated AIGC workload scheduling framework, which introduces an explicit mathematical characterization of service quality to promote both job transfer among ASPs and fine-grained inference process configuration. Moreover, various energy resources within data centers are jointly considered to enhance power usage flexibility. Subsequently, a system utility maximization problem is formulated to balance AIGC service revenue with operational penalties and costs. Nevertheless, the strong coupling among job scheduling decisions induces severe reward sparsity, which limits the effectiveness of existing deep reinforcement learning (DRL) algorithms. To address this issue, we develop a diffusion model-aided reward shaping approach to synthesize complementary reward signals through a multi-step denoising process. This approach is seamlessly integrated with DRL to enable efficient learning of scheduling policies under sparse environmental feedback. Experiments based on real-world models and datasets demonstrate that our scheme effectively accommodates electricity price fluctuations and AIGC model heterogeneity, while achieving superior learning convergence and system utility compared with benchmark methods.
Yang Fu, Peng Qin, Liming Chen +3
Yang Fu, Peng Qin, Member, IEEE, Liming Chen, Zihao Zhang, Hao Yu, and Yifei Wang