Large language models (LLMs) are emerging as key enablers of automation in domains such as telecommunications, assisting with tasks including troubleshooting, standards interpretation, and network optimization. However, their deployment in practice must balance inference cost, latency, and reliability. In this work, we study an edge-cloud-expert cascaded LLM-based knowledge system that supports decision-making through a question-and-answer pipeline. In it, an efficient edge model handles routine queries, a more capable cloud model addresses complex cases, and human experts are involved only when necessary. We define a misalignment-cost constrained optimization problem, aiming to minimize average processing cost, while guaranteeing alignment of automated answers with expert judgments. We propose a statistically rigorous threshold selection method based on multiple hypothesis testing (MHT) for a query processing mechanism based on knowledge and confidence tests. The approach provides finite-sample guarantees on misalignment risk. Experiments on the TeleQnA dataset –a telecom-specific benchmark – demonstrate that the proposed method achieves superior cost-efficiency compared to conventional cascaded baselines, while ensuring reliability at prescribed confidence levels.

In modern wireless network architectures, such as O-RAN, artificial intelligence (AI)-based applications are deployed at intelligent controllers to carry out functionalities like scheduling or power control. The AI “apps” are selected on the basis of contextual information such as network conditions, topology, traffic statistics, and design goals. The mapping between context and AI model parameters is ideally done in a zero-shot fashion via an automatic model selection (AMS) mapping that leverages only contextual information without requiring any current data. This paper introduces a general methodology for the online optimization of AMS mappings. Optimizing an AMS mapping is challenging, as it requires exposure to data collected from many different contexts. Therefore, if carried out online, this initial optimization phase would be extremely time consuming. A possible solution is to leverage a digital twin of the physical system to generate synthetic data from multiple simulated contexts. However, given that the simulator at the digital twin is imperfect, a direct use of simulated data for the optimization of the AMS mapping would yield poor performance when tested in the real system. This paper proposes a novel method for the online optimization of AMS mapping that corrects for the bias of the simulator by means of limited real data collected from the physical system. Experimental results for a graph neural network-based power control app demonstrate the significant advantages of the proposed approach.

In recent years, the application of neural networks (NNs) in wireless communication has garnered widespread attention and proven successful. However, conventional learning-based NNs often suffer from poor calibration, meaning that they struggle to reliably quantify prediction confidence and lack proper uncertainty estimation. This limitation becomes especially critical for next generation communication systems, particularly in complex industrial scenarios with stringent reliability requirements. Previous efforts to enhance model calibration have primarily centered on modifying NNs’ training processes. However, these methods often demand significant computing resources, making them impractical for resource-constrained scenarios. In this paper, we investigate a distributed wireless communication system involving multiple users and propose a novel approach to improve model calibration. Our method focuses on enhancing calibration during the inference stage of NNs by introducing a power control mechanism. Notably, existing research indicates that many NNs exhibit overconfidence, i.e., the NN’s confidence exceeds its actual accuracy. Leveraging this insight, we exploit the inherent noise and fading in wireless systems to naturally reduce the NN’s confidence while preserving accuracy. To achieve this, we employ linear relaxation-based perturbation analysis (LiRPA) to approximate the relationship between the perturbed output and the input perturbation of the NN. Subsequently, we devise an optimization problem by leveraging the analyzed relationship and the definition of perfect calibration. It is aimed at finding the input perturbation that maximizes the probability of the model achieving perfect calibration. Finally, considering different channel conditions and a given specific modulation method, we derive the optimal transmission power based on bit error rate (BER) formula. Simulation results demonstrate that our proposed power control method exhibits significant advantages in model calibration compared to several traditional approaches.

With the great success of deep learning (DL) in image classification, speech recognition, and other fields, more and more studies have applied various neural networks (NNs) to wireless resource allocation. Generally speaking, these artificial intelligent (AI) models are trained under some special learning hypotheses, especially that the statistics of the training data are static during the training stage. However, the distribution of channel state information (CSI) is constantly changing in the real-world wireless communication environment. Therefore, it is essential to study effective dynamic DL technologies to solve wireless resource allocation problems. In this paper, we propose a novel framework, named meta-gating, for solving resource allocation problems in an episodically dynamic wireless environment, where the CSI distribution changes over periods and remains constant within each period. The proposed framework, consisting of an inner network and an outer network, aims to adapt to the dynamic wireless environment by achieving three important goals, i.e., seamlessness, quickness and continuity. Specifically, for the former two goals, we propose a training method by combining a model-agnostic meta-learning (MAML) algorithm with an unsupervised learning mechanism. With this training method, the inner network is able to fast adapt to different channel distributions because of the good initialization. As for the goal of ‘continuity’, the outer network can learn to evaluate the importance of inner network’s parameters under different CSI distributions, and then decide which subset of the inner network should be activated through the gating operation. Additionally, we theoretically analyze the performance of the proposed meta-gating framework. Simulation results demonstrate that the proposed meta-gating framework can well achieve the three important goals compared with existing state-of-the-art algorithms.

As an important part of the fifth generation (5G) mobile networks, unmanned aerial vehicles (UAVs) have been applied in various communication scenarios due to their high operability and low cost. In this paper, we investigate a multi-UAV communication system with moving users and consider the co-channel interference caused by the transmissions of all other UAVs. To ensure the fairness, we maximize the minimum average user rate during the observed time by jointly optimizing UAVs’ trajectories, transmission power, and user association. Considering that UAVs can cover a large area for communications, UAVs do not need to move as soon as the users move. Therefore, a two-timescale structure is proposed for the considered scenario, where the UAVs’ trajectories are optimized based on the channel state information (CSI) in a long timescale, while the transmission power and the user association are optimized based on the instantaneous CSI in a short timescale. To effectively tackle this challenging non-convex problem with both discrete and continuous variables, we propose a joint neural network (NN) design, where a deep reinforcement learning based Pointer Network named advantage pointer-critic (APC) is applied to optimize discrete variables and a deep-unfolding NN is used to optimize the continuous variables. Specifically, we first formulate a Markov decision process to model the user association, and then employ the APC network trained by the advantage actor-critic algorithm to address it. The APC network consists of a Pointer Network and a Multilayer Perceptron. As for the deep-unfolding NN, we first develop a block coordinate descent based algorithm to optimize the UAVs’ trajectories and transmission power, and then unfold the algorithm into a layer-wise NN with introduced trainable parameters. These two networks are jointly trained in an unsupervised fashion. Simulation results validate that the proposed joint NN significantly outperforms the optimization algorithm with much lower complexity, and achieves good performances on scalability and generalization ability.

In this paper, we investigate a multi-UAV communication system with moving users and consider the co-channel interference caused by the transmissions of all other UAVs. To ensure the fairness of moving users, we maximize the minimum average user rate during the observed time by jointly optimizing UAVs' trajectories, transmission power, and user association. To effectively tackle this non-convex problem with both discrete and continuous variables, we propose a joint neural network (NN) design, where a network named advantage pointer-critic (APC) is applied to optimize discrete variables and a deep-unfolding NN is used to optimize continuous variables. Specifically, we first elaborately formulate a Markov decision process to model the user association, and then use the APC network trained by the advantage actor-critic algorithm to address it. As for the deep-unfolding NN, we first develop a block coordinate descent based algorithm to optimize UAVs' trajectories and transmission power, and then unfold this algorithm into a layer-wise NN with introduced trainable parameters. These two networks are jointly trained in an unsupervised fashion. Simulation results validate that the proposed joint NN significantly outperforms the mathematical optimization algorithm with much lower complexity.