Fine-Grained AI Model Caching and Downloading With Coordinated Multipoint Broadcasting in Multi-Cell Edge Networks

TL;DR

This work tackles the latency challenges of on-demand AI model downloading in multi-cell edge networks by proposing a fine-grained caching scheme that stores parameter blocks (PBs) and leverages coordinated multipoint (CoMP) broadcasting. It introduces MAASN-DA, a distributed multi-agent deep reinforcement learning framework that captures action semantics, uses data augmentation with an echo state network, and employs a value-decomposition critic to jointly optimize PB caching, PB migration, and CoMP beamforming under CSI uncertainty. A robust CoMP beamforming subroutine based on S-Procedure and DC programming ensures QoS requirements are met despite imperfect CSI. Theoretical convergence guarantees and extensive simulations demonstrate significant reductions in model downloading delay (up to about 67.86%) and validate the two-fold gains from caching efficiency and simultaneous PB broadcasting, with extensions to large-language-model (LLM) and Llama-based caching scenarios showing strong scalability and robustness.

Abstract

6G networks are envisioned to support on-demand AI model downloading to accommodate diverse inference requirements of end users. By proactively caching models at edge nodes, users can retrieve the requested models with low latency for on-device AI inference. However, the substantial size of contemporary AI models poses significant challenges for edge caching under limited storage capacity, as well as for the concurrent delivery of heterogeneous models over wireless channels. To address these challenges, we propose a fine-grained AI model caching and downloading system that exploits parameter reusability, stemming from the common practice of fine-tuning task-specific models from a shared pre-trained model with frozen parameters. This system selectively caches model parameter blocks (PBs) at edge nodes, eliminating redundant storage of reusable parameters across different cached models. Additionally, it incorporates coordinated multipoint (CoMP) broadcasting to simultaneously deliver reusable PBs to multiple users, thereby enhancing downlink spectrum utilization. Under this arrangement, we formulate a model downloading delay minimization problem to jointly optimize PB caching, migration (among edge nodes), and broadcasting beamforming. To tackle this intractable problem, we develop a distributed multi-agent learning framework that enables edge nodes to explicitly learn mutual influence among their actions, thereby facilitating cooperation. Furthermore, a data augmentation approach is proposed to adaptively generate synthetic training samples through a predictive model, boosting sample efficiency and accelerating policy learning. Both theoretical analysis and simulation experiments validate the superior convergence performance of the proposed learning framework.

Figures (14)

• Figure 1: Fine-grained AI model caching and downloading system model.
• Figure 2: Identification of PBs across AI model architectures.
• Figure 3: The proposed MAASN-DA framework.
• Figure 4:
• Figure 6: Upper-bound of Q-value approximation error versus $\tau_0$ and $\xi$.