EdgeCoInfer, a hierarchical collaborative inference framework enabling efficient on-device MLLM inference through coarse-to-fine orchestration, and a Hybrid Evolutionary Hierarchical Reinforcement Learning (HE-HRL) framework to tackle the non-differentiable combinatorial explosion.
Abstract
To deliver ubiquitous intelligence, modern mobile applications increasingly execute concurrent Multimodal Large Language Models (MLLMs) on edge devices, presenting severe challenges under multi-task concurrency and tight resource constraints. To address this, we propose EdgeCoInfer, a hierarchical collaborative inference framework enabling efficient on-device MLLM inference through coarse-to-fine orchestration. Coarsely, EdgeCoInfer decomposes MLLMs into functional modules for inter-task sharing, avoiding redundant model loading. Finely, it partitions models at the neural network layer level and distributes segments across devices and servers. We jointly optimize layer partitioning, module sharing, and resource allocation under tight constraints. To tackle the non-differentiable combinatorial explosion, we propose a Hybrid Evolutionary Hierarchical Reinforcement Learning (HE-HRL) framework. HE-HRL synchronizes a gradient-free genetic algorithm for discrete partitioning and sharing decisions with a gradient-based soft actor-critic agent for continuous resource refinement. We further embed a constructive cut-step decoder with pre-act pruning and a two-phase curriculum to improve feasibility and accelerate convergence. Experimental results show that EdgeCoInfer breaks the edge memory wall and prevents catastrophic out-of-memory and task failures under high concurrency, reducing memory demand by 53.53\% and system cost by 59.86\% compared to existing methods.
Collaborative inference pools distributed resources to run compute-intensive Vision Transformers (ViTs) in satellite edge computing. Model partitioning enables such collaboration by assigning consecutive layer groups to different nodes, but the large volume of intermediate activation data incurs substantial transfer overhead that can erase its benefit. Token compression reduces downstream computation and activation transfer, but its quality impact depends on input content, model depth, and earlier pruning decisions, while layer offloading must adapt to time-varying contact and battery conditions. We present \sys, a content-aware hierarchical scheduler that screens constellation-wide options to retain a bounded candidate set, then refines each candidate into a complete token compression and layer offloading trajectory using quality prediction and joint planning. A unified objective balances per-task latency, energy, and quality loss against accumulated workload and battery pressures. We implement \sys on an NVIDIA Jetson AGX Orin hardware-in-the-loop testbed and use its validated execution model for constellation-scale trace replay across multiple ViT workloads and constellation settings. At \(5\)~tasks/s, \sys accomplishes 91.6\% of released tasks, 26.1 percentage points above MARATD3, the strongest baseline, while reducing mean latency and battery draw by 53.0\% and 70.8\%, respectively, and meeting quality targets.
Yan Chen, Yun-Xiang Zhang, Guan-Jun Jiang et al.· 0 citations
6G mobile edge networks are emerging as a key infrastructure for ubiquitous large language model (LLM) inference services. However, conventional edge routing to nearby or well-connected servers falls short for efficient edge LLM inference, as it may miss the user’s KV cache and trigger costly prefill recomputation. To address this challenge, this paper studies an edge inference system assisted by an embodied UAV agent swarm, where UAVs actively sense user mobility and neighboring UAV states to make local decisions on trajectory control, user association, and inference-request routing. The goal is to improve KV-cache reuse while maintaining reliable wireless connectivity, thereby maximizing the system effective token throughput under energy and QoS constraints. We then formulate the joint optimization as a mixed-integer non-linear program and further cast the sequential UAV decision-making process as a decentralized partially observable Markov decision process. To obtain scalable decentralized policies under partial observations, we propose Q-MAA2C, a quantum-enhanced multi-agent advantage actor-critic algorithm for embodied UAV swarm control and inference routing. Q-MAA2C uses quantum actors for local action selection and an entangled split critic for swarm-level value estimation, enabling coordinated policies from partial observations with reduced raw observation exchange. Simulation results indicate that Q-MAA2C yields comparable reinforcement learning rewards to the fully classical baseline while reducing the number of convergence episodes by about 43%. Additionally, the proposed method enhances the system effective token throughput by up to about 134% over other competing methods.
Xiangdong Zheng, Long Luo, Hongfang Yu et al.· IEEE Transactions on Cogniti...· 1 citation
Enabling collaborative generative AI (GenAI) inference at the network edge is challenging due to limited caching capacity, heterogeneous computing resources, and highly dynamic, latency-sensitive service demands. In this paper, we investigate the joint optimization of GenAI model caching, inference offloading, and resource allocation in a collaborative cloud–edge–end architecture. To address the strong coupling between long-term caching decisions and short-term scheduling dynamics, we propose a Hierarchical Meta-Graph Reinforcement Learning framework, termed HMGRL. Specifically, a heat-greedy model caching strategy is developed to capture time-varying model popularity and to reduce switching overhead on a slow timescale, while a graph-enhanced dueling deep reinforcement learning algorithm with prioritized experience replay enables topology-aware collaborative inference offloading and resource allocation on a fast timescale. Extensive simulations demonstrate that HMGRL consistently outperforms representative baselines in terms of system utility, cache and computing-resource utilization, convergence stability, and performance robustness. These results validate the effectiveness of the proposed hierarchical learning framework for practical GenAI applications at the network edge.
Liang Zhao, Jing Wei, Huan Zhou et al.· IEEE Transactions on Cogniti...· 0 citations
An inference-aware AirMoE error metric is constructed to quantify aggregation distortion effects on end-to-end (E2E) inference accuracy via perturbation-based layer-sensitivity calibration, and an activation- and channel-aware expert placement strategy is developed that assigns more important experts to devices with lower channel-power cost.
The increasing scale and computational demands of large artificial intelligence models (LAIMs) present significant challenges for efficient inference in resource-constrained distributed environments. In this paper, we propose a multi-cluster LAIM co-inference framework, where an edge server equipped with multiple graphics processing units (GPUs) coordinates multiple user clusters to execute inference tasks collaboratively. Within each cluster, devices capture data from diverse perspectives and employ lightweight on-device LAIMs to extract local features. These features are then transmitted to the edge server, where they are aggregated and fused to generate a more accurate inference outcome. To reveal the fundamental trade-off between model pruning and collaborative inference performance, we develop a theoretical framework that characterizes the impact of pruning ratios and device contributions using rate-distortion theory and partial information decomposition. Based on this analysis, we formulate a joint optimization problem that determines the model pruning ratio, the task scheduling strategy, the bandwidth allocation, and the transmission power, with the goal of minimizing the inference distortion while satisfying the constraints of latency, energy consumption, and server capacity. Extensive simulation results demonstrate that the proposed framework significantly outperforms existing benchmark schemes, achieving superior inference accuracy and resource efficiency in multi-cluster edge intelligence networks.
Xiaowen Cao, Zhonghao Lyu, Shicheng Chu et al.· 0 citations
The evolving RAN intelligent controller (RIC) (EvoRIC) framework is introduced, a hierarchical architecture that enables continuous evolution by leveraging a non-real-time RIC (non-RT RIC) for global model updates and a near-real-time RIC (near-RT RIC) for local execution, dynamically empowering LLMs with domain-specific decision-making capabilities.
Lingyan Bao, Jemin Lee, Tony Q. S. Quek· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
