Adaptive RAN Slicing Control via Reward-Free Self-Finetuning Agents

2026-03-11

AI Summary

This work addresses the challenges generative AI models face in continuous control tasks within AI-native networks—namely limited context windows, absence of explicit rewards, and performance degradation over long horizons. The authors propose a self-finetuning framework that operates without external rewards: leveraging a dual-perspective reflection mechanism, the agent generates linguistic feedback from its interaction history to construct a preference dataset, which is then used in preference-based learning to internalize long-term experience into its parameters. This approach establishes a novel reward-free paradigm for continuous control grounded in self-reflection and preference learning, enabling generative agents to autonomously refine their policies and overcome limitations of both conventional reinforcement learning and prompt-dependent LLM agents. Evaluated on dynamic wireless network slicing, the method significantly outperforms baselines in sample efficiency, control stability, and multi-objective optimization across spectral efficiency, quality of service, and configuration stability.

Technology Category

• Natural Language Processing: Generation
• Humans and AI: Learning Human Values and Preferences
• Machine Learning: Life-Long and Continual Learning

Application Category

• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Search and Retrieval-Augmented AI: Agentic search

Abstract

The integration of Generative AI models into AI-native network systems offers a transformative path toward achieving autonomous and adaptive control. However, the application of such models to continuous control tasks is impeded by intrinsic architectural limitations, including finite context windows, the lack of explicit reward signals, and the degradation of the long context. This paper posits that the key to unlocking robust continuous control is enabling agents to internalize experience by distilling it into their parameters, rather than relying on prompt-based memory. To this end, we propose a novel self-finetuning framework that enables agentic systems to learn continuously through direct interaction with the environment, bypassing the need for handcrafted rewards. Our framework implements a bi-perspective reflection mechanism that generates autonomous linguistic feedback to construct preference datasets from interaction history. A subsequent preference-based fine-tuning process distills long-horizon experiences into the model's parameters. We evaluate our approach on a dynamic Radio Access Network (RAN) slicing task, a challenging multi-objective control problem that requires the resolution of acute trade-offs between spectrum efficiency, service quality, and reconfiguration stability under volatile network conditions. Experimental results show that our framework outperforms standard Reinforcement Learning (RL) baselines and existing Large Language Model (LLM)-based agents in sample efficiency, stability, and multi-metric optimization. These findings demonstrate the potential of self-improving generative agents for continuous control tasks, paving the way for future AI-native network infrastructure.

Problem

• RAN slicing
• continuous control
• reward-free learning
• multi-objective optimization
• AI-native networks

Innovation

• self-finetuning
• reward-free learning
• bi-perspective reflection
• RAN slicing
• preference-based fine-tuning