LAMBO: Large AI Model Empowered Edge Intelligence

TL;DR

The paper tackles offloading decisions in multi-access edge computing under heterogeneous constraints and dynamic environments. It proposes LAMBO, a Large AI Model-Based Offloading framework with over one billion parameters, featuring input embedding, an asymmetric encoder-decoder, actor-critic pre-training, and expert-feedback-based fine-tuning guided by task prompts. Key contributions include a high-quality representation of heterogeneous constraints via input embeddings, a global-decision AED architecture, multi-task generalization through ACL, and adaptive fine-tuning via ALEF for changing environments; simulations demonstrate that larger LAMBO models achieve superior latency and energy performance. The work highlights the potential of deploying large AI models at the network edge to deliver robust, scalable, and adaptable MEC services, while acknowledging open issues in memory, compute, and communication overhead that warrant further research.

Abstract

Next-generation edge intelligence is anticipated to benefit various applications via offloading techniques. However, traditional offloading architectures face several issues, including heterogeneous constraints, partial perception, uncertain generalization, and lack of tractability. In this paper, we propose a Large AI Model-Based Offloading (LAMBO) framework with over one billion parameters for solving these problems. We first use input embedding (IE) to achieve normalized feature representation with heterogeneous constraints and task prompts. Then, we introduce a novel asymmetric encoder-decoder (AED) as the decision-making model, which is an improved transformer architecture consisting of a deep encoder and a shallow decoder for global perception and decision. Next, actor-critic learning (ACL) is used to pre-train the AED for different optimization tasks under corresponding prompts, enhancing the AED's generalization in multi-task scenarios. Finally, we propose an active learning from expert feedback (ALEF) method to fine-tune the decoder of the AED for tracking changes in dynamic environments. Our simulation results validate the advantages of the proposed LAMBO framework.

Figures (6)

• Figure 1: Deep offloading architecture versus LAMBO architecture.
• Figure 2: Structures of IE and AED.
• Figure 3: Workflow of ACL.
• Figure 4: Workflow of ALEF.
• Figure 5: Comparison of different schemes in task latency.