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.

We introduce adaptive learn-then-test (aLTT), an efficient hyperparameter selection procedure that provides finite-sample statistical guarantees on the population risk of AI models. Unlike the existing learn-then-test (LTT) technique, which relies on conventional p-value-based multiple hypothesis testing (MHT), aLTT implements sequential data-dependent MHT with early termination by leveraging e-processes. As a result, aLTT can reduce the number of testing rounds, making it particularly well-suited for scenarios in which testing is costly or presents safety risks. Apart from maintaining statistical validity, in applications such as online policy selection for offline reinforcement learning and prompt engineering, aLTT is shown to achieve the same performance as LTT while requiring only a fraction of the testing rounds.

We investigate a lossy source compression problem in which both the encoder and decoder are equipped with a pre-trained sequence predictor. We propose an online lossy compression scheme that, under a$0-1$loss distortion function, ensures a deterministic, per-sequence upper bound on the distortion (outage) level for any time instant. The outage guarantees apply irrespective of any assumption on the distribution of the sequences to be encoded or on the quality of the predictor at the encoder and decoder. The proposed method, referred to as online conformal compression (OCC), is built upon online conformal prediction-a novel method for constructing confidence intervals for arbitrary predictors. Numerical results show that OCC achieves a compression rate comparable to that of an idealized scheme in which the encoder, with hindsight, selects the optimal subset of symbols to describe to the decoder, while satisfying the overall outage constraint.

Predictive models are often required to produce reliable predictions under statistical conditions that are not matched to the training data. A common type of training-testing mismatch is covariate shift, where the conditional distribution of the target variable given the input features remains fixed, while the marginal distribution of the inputs changes. Weighted conformal risk control (W-CRC) uses data collected during the training phase to convert point predictions into prediction sets with valid risk guarantees at test time despite the presence of a covariate shift. However, while W-CRC provides statistical reliability, its efficiency - measured by the size of the prediction sets - can only be assessed at test time. In this work, we relate the generalization properties of the base predictor to the efficiency of W-CRC under covariate shifts. Specifically, we derive a bound on the inefficiency of the W-CRC predictor that depends on algorithmic hyperparameters and task-specific quantities available at training time. This bound offers insights on relationships between the informativeness of the prediction sets, the extent of the covariate shift, and the size of the calibration and training sets. Experiments on fingerprinting-based localization validate the theoretical results.

Selecting artificial intelligence (AI) models, such as large language models (LLMs), from multiple candidates requires accurate performance estimation. This is ideally achieved through empirical evaluations involving abundant real-world data. However, such evaluations are costly and impractical at scale. To address this challenge, autoevaluation methods leverage synthetic data produced by automated evaluators, such as LLMs-as-judges, reducing variance but potentially introducing bias. Recent approaches have employed semi-supervised prediction-powered inference ($\texttt{PPI}$) to correct for the bias of autoevaluators. However, the use of autoevaluators may lead in practice to a degradation in sample efficiency compared to conventional methods using only real-world data. In this paper, we propose $\texttt{R-AutoEval+}$, a novel framework that provides finite-sample reliability guarantees on the model evaluation, while also ensuring an enhanced (or at least no worse) sample efficiency compared to conventional methods. The key innovation of $\texttt{R-AutoEval+}$ is an adaptive construction of the model evaluation variable, which dynamically tunes its reliance on synthetic data, reverting to conventional methods when the autoevaluator is insufficiently accurate. Experiments on the use of LLMs-as-judges for the optimization of quantization settings for the weights of an LLM, for prompt design in LLMs, and for test-time reasoning budget allocation in LLMs confirm the reliability and efficiency of $\texttt{R-AutoEval+}$.

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.

The safe integration of machine learning modules in decision-making processes hinges on their ability to quantify uncertainty. A popular technique to achieve this goal is conformal prediction (CP), which transforms an arbitrary base predictor into a set predictor with coverage guarantees. While CP certifies the predicted set to contain the target quantity with a user-defined tolerance, it does not provide control over the average size of the predicted sets, i.e., over the informativeness of the prediction. In this work, a theoretical connection is established between the generalization properties of the base predictor and the informativeness of the resulting CP prediction sets. To this end, an upper bound is derived on the expected size of the CP set predictor that builds on generalization error bounds for the base predictor. The derived upper bound provides insights into the dependence of the average size of the CP set predictor on the amount of calibration data, the target reliability, and the generalization performance of the base predictor. The theoretical insights are validated using simple numerical regression and classification tasks.

Adaptive Risk Control (ARC) is an online calibration strategy based on set prediction that offers worst-case deterministic long-term risk control, as well as statistical marginal coverage guarantees. ARC adjusts the size of the prediction set by varying a single scalar threshold based on feedback from past decisions. In this work, we introduce Localized Adaptive Risk Control (L-ARC), an online calibration scheme that targets statistical localized risk guarantees ranging from conditional risk to marginal risk, while preserving the worst-case performance of ARC. L-ARC updates a threshold function within a reproducing kernel Hilbert space (RKHS), with the kernel determining the level of localization of the statistical risk guarantee. The theoretical results highlight a trade-off between localization of the statistical risk and convergence speed to the long-term risk target. Thanks to localization, L-ARC is demonstrated via experiments to produce prediction sets with risk guarantees across different data subpopulations, significantly improving the fairness of the calibrated model for tasks such as image segmentation and beam selection in wireless networks.

Standard Bayesian learning is known to have suboptimal generalization capabilities under misspecification and in the presence of outliers. Probably approximately correct (PAC)-Bayes theory demonstrates that the free energy criterion minimized by Bayesian learning is a bound on the generalization error for Gibbs predictors (i.e., for single models drawn at random from the posterior) under the assumption of sampling distributions uncontaminated by outliers. This viewpoint provides a justification for the limitations of Bayesian learning when the model is misspecified, requiring ensembling, and when data are affected by outliers. In recent work, PAC-Bayes bounds-referred to as PACm-were derived to introduce free energy metrics that account for the performance of ensemble predictors, obtaining enhanced performance under misspecification. This work presents a novel robust free energy criterion that combines the generalized logarithm score function with PACm ensemble bounds. The proposed free energy training criterion produces predictive distributions that are able to concurrently counteract the detrimental effects of misspecification-with respect to both likelihood and prior distribution-and outliers.

Decentralized learning enables edge users to collaboratively train models by exchanging information via device-to-device communication, yet prior works have been limited to wireless networks with fixed topologies and reliable workers. In this work, we propose an asynchronous decentralized stochastic gradient descent (DSGD) algorithm, which is robust to the inherent computation and communication failures occurring at the wireless network edge. We theoretically analyze its performance and establish a non-asymptotic convergence guarantee. Experimental results corroborate our analysis, demonstrating the benefits of asynchronicity and outdated gradient information reuse in decentralized learning over unreliable wireless networks.