IGAA: Intent-Driven General Agentic AI for Edge Services Scheduling using Generative Meta Learning

TL;DR

IGAA addresses the problem of generalizing agentic-edge service scheduling across heterogeneous tasks and mobility-driven dynamics. It introduces a generative meta-learning framework driven by an intent-centric N–S–I mapping that grounds high-level user intents into numerical resource demands, enabling scalable dataset generation and policy synthesis. The core contributions are RCETL and APOTL for easy-to-hard adaptation, a Generative Intent Replay mechanism to prevent forgetting, and a scenario evaluation-correction loop to mitigate LLM hallucinations, all within an IGAA meta-learning loop. Empirical results show strong generalization and scalability across diverse edge scenarios, with transfer of latency-sensitive patterns across domains and the intent-satisfaction gap remaining within 3.81%.

Abstract

Agentic AI (AAI), which extends Large Language Models with enhanced reasoning capabilities, has emerged as a promising paradigm for autonomous edge service scheduling. However, user mobility creates highly dynamic service demands in edge networks, and existing service scheduling agents often lack generalization capabilities for new scenarios. Therefore, this paper proposes a novel Intent-Driven General Agentic AI (IGAA) framework. Leveraging a meta-learning paradigm, IGAA enables AAI to continuously learn from prior service scheduling experiences to achieve generalized scheduling capabilities. Particularly, IGAA incorporates three core mechanisms. First, we design a Network-Service-Intent matrix mapping method to allow agents to simulate novel scenarios and generate training datasets. Second, we present an easy-to-hard generalization learning scheme with two customized algorithms, namely Resource Causal Effect-aware Transfer Learning (RCETL) and Action Potential Optimality-aware Transfer Learning (APOTL). These algorithms help IGAA adapt to new scenarios. Furthermore, to prevent catastrophic forgetting during continual IGAA learning, we propose a Generative Intent Replay (GIR) mechanism that synthesizes historical service data to consolidate prior capabilities. Finally, to mitigate the effect of LLM hallucinations on scenario simulation, we incorporate a scenario evaluation and correction model to guide agents in generating rational scenarios and datasets. Extensive experiments demonstrate IGAA's strong generalization and scalability. Specifically, IGAA enables rapid adaptation by transferring learned policies to analogous new ones, such as applying latency-sensitive patterns from real-time computing to optimize novel Internet of Vehicles (IoV) services. Compared to scenario-specific methods, IGAA maintains the intent-satisfaction rate gap within 3.81%.

Figures (10)

• Figure 1: Overview of IGAA (inference phase). IGAA transforms the natural language intent proposed by users into intent vectors and maps them into resource requirements vectors. Then, the service scheduling model deployed in IGAA generates personalized service scheduling policies accordingly.
• Figure 2: Overview of IGAA (training phase). IGAA simulates new service scenarios according to CoT and updates the N-S-I matrix using the proposed algorithm. Then, IGAA generates new training code and datasets to enable the service scheduling model to acquire scheduling capabilities in the new scenario. Finally, the scheduling model performs transfer learning using the generated code and datasets. Besides, the scheduling model continuously interacts with the LLM to optimize the scheduling performance.
• Figure 3: RCETL aims to perceive the impact of newly introduced resources on existing ones and selectively fine-tune the parameters that are strongly correlated with the new resources. In APOTL, different sub-networks leverage prior service scheduling experience to optimize scheduling strategies when new service types are introduced. We derive the upper bounds of sub-network action potential optimality and use them to refine the service scheduling behavior selection strategy.
• Figure 4: Overview of Generative Intent Replay. IGAA samples earlier simulated service scenarios and generates the corresponding service intents. These intents are then combined with those from new scenarios to form new dataset.
• Figure 5: Scalability analysis of the IGAA framework. Even when $R$ = 10 and $S$ = 10, both intent satisfaction rate and sucess rate remain above 90%