Towards Adaptive Networks - Generalized utility functions in Multi-Agent Frameworks

Autonomous management of multiple services is a crucial requirement of 6G networks. In recent years, researchers have been working on using artificial intelligence(AI) to orchestrate intents, autonomously, through Intent Management Functions (IMFs). These IMFs can handle conflicting service intents and prioritize the global objective based on predefined utility function and intent priorities. However, for such frameworks to be successful in real-life scenarios, they must be flexible to business situations. Service priorities can change, and the utility function that measures the fulfillment of objectives may also vary in definition. This paper proposes a novel method that enables the IMF to adapt to unseen forms of utility functions and changes in service priorities at run-time without requiring additional training. We assume the IMF contains agents trained using multi-agent reinforcement learning (MARL) to perform actions and multiple such MARL systems are orchestrated by Ad-hoc teaming approaches. Results on a network emulator demonstrate the effectiveness of the approach, outperforming existing state-of-the-art methods that require additional training to achieve the same flexibility, thereby saving costs and increasing adaptability.