In this paper, we study the problem of federated learning over a wireless channel with user sampling, modeled by a fading multiple access channel, subject to central and local differential privacy (DP/LDP) constraints. It has been shown that the superposition nature of the wireless channel provides a dual benefit of bandwidth efficient gradient aggregation, in conjunction with strong DP guarantees for the users. Specifically, the central DP privacy leakage has been shown to scale as $\mathcal {O}(1/K^{1/2})$ , where $K$ is the number of users. It has also been shown that user sampling coupled with orthogonal transmission can enhance the central DP privacy leakage with the same scaling behavior. In this work, we show that, by jointly incorporating both wireless aggregation and user sampling, one can obtain even stronger privacy guarantees. We propose a private wireless gradient aggregation scheme, which relies on independently randomized participation decisions by each user. The central DP leakage of our proposed scheme scales as $\mathcal {O}(1/K^{3/4})$ . In addition, we show that LDP is also boosted by user sampling. We also present analysis for the convergence rate of the proposed scheme and study the tradeoffs between wireless resources, convergence, and privacy theoretically and empirically for two scenarios when the number of sampled participants are $(a)$ known, or $(b)$ unknown at the parameter server.

In this paper, we study Local Information Privacy (LIP). As a context-aware privacy notion, LIP relaxes the de facto standard privacy notion of local differential privacy (LDP) by incorporating prior knowledge and therefore achieving better utility. We study the relationships between LIP and some of the representative privacy notions including LDP, mutual information and maximal leakage. We show that LIP provides strong instance-wise privacy protection compared to other context-aware privacy notions. Moreover, we present some useful properties of LIP, including post-processing, linkage, composability, transferability and robustness to imperfect prior knowledge. Then we study a general utility-privacy tradeoff framework, under which we derive LIP based privacy-preserving mechanisms for both discrete and continuous-valued data. Three types of perturbation mechanisms are studied in this paper: 1) randomized response (RR), 2) random sampling (RS) and 3) additive noise (AN) (e.g., Gaussian mechanism). Our privacy mechanisms incorporate the prior knowledge into the perturbation parameters so as to enhance utility. Finally, we present a comprehensive set of experiments on real datasets to illustrate the advantage of context-awareness and compare the utility-privacy tradeoffs provided by different mechanisms.

Split Federated Learning (SFL) has recently emerged as a promising distributed learning technology that leverages the strengths of both federated and split learning. In this approach, clients are responsible for training only part of the model, termed the client-side model, thereby alleviating their computational burden. Then, clients can enhance their convergence speed by synchronizing these client-side models. Consequently, SFL has received significant attention from both industry and academia, with diverse applications in 6G networks. However, while offering considerable benefits, SFL introduces additional communication overhead when interacting with servers. Moreover, the current SFL method presents several privacy concerns during frequent interactions. In this context, the choice of the cut layer in SFL, which splits the model into both client- and server-side models, can substantially impact the energy consumption of clients and their privacy because it influences the training burden and output of the client-side models. Correspondingly, extensive research is required to analyze the impact of cut layer selection, and careful consideration should be given to this aspect. Therefore, this study provides a comprehensive overview of the SFL process and reviews its state-of-the-art. We thoroughly analyze the energy consumption and privacy regarding the cut layer selection in SFL by considering the influence of various system parameters on the selection strategy. Moreover, we provide an illustrative example of cut layer selection to minimize the clients’ risk of reconstructing raw data at the server. This is done while sustaining energy consumption within the required energy budget, which involves trade-offs. We also discuss other control variables that can be optimized in conjunction with the cut layer selection. Finally, we highlight open challenges in this field. These are promising avenues for future research and development.

In this letter, a novel cut layer selection scheme is designed to minimize the overall latency in split federated learning (SFL) over wireless networks, while maintaining an acceptable privacy level. Our approach considers the cut layer selection management, which determines the split point that partitions the entire model into two distinct sub-models: client-side and server-side models. Thus, we develop analytical models to assess both the overall latency and privacy level throughout the SFL process, recognizing the pivotal role of cut layer selection in influencing the training burden and output of client-side models, which pose a risk of data reconstruction by adversaries (i.e., reconstruction attack). Considering a tradeoff between overall latency and privacy level in terms of the cut layer selection, we establish a theoretical framework for managing cut layer selection in SFL to optimize the cut layer point. Furthermore, we discuss the impact of a differential privacy technique designed to enhance privacy by effectively concealing individual information. We evaluate the performance of the proposed scheme and provide insights on optimizing the overall latency of SFL while maintaining the desired privacy level through cut layer selection.

Hybrid language models (HLMs) are inference-time architectures that combine the low-latency efficiency of small language models (SLMs) on clients (edge devices) with the high accuracy of large language models (LLMs) in centralized servers. Unlike traditional end-to-end LLM inference, HLMs aim to reduce latency and communication by selectively invoking LLMs only when the local SLM’s predictions are uncertain—that is, when the model exhibits low confidence or high entropy in its token-level probability distribution. However, when the SLM encounters ambiguous or low-confidence predictions during inference, it must offload token-level probability distributions to the LLM for refinement. This frequent offloading leads to substantial communication overhead, particularly in bandwidth-constrained environments. To address this challenge, we propose federated learning (FL)-enabled HLM (FedHLM), a communication-efficient HLM framework that integrates uncertainty-aware inference with FL. The key innovation lies in collaboratively learning token-level uncertainty thresholds that determine when SLM predictions require LLM assistance. Instead of relying on static or hand-tuned thresholds, FedHLM uses FL to enable distributed threshold optimization across clients while preserving data privacy. Additionally, embedding-based token representations are employed to facilitate semantic similarity comparisons during peer-to-peer (P2P) resolution, allowing clients to reuse tokens inferred by similar peers without efficiently involving the LLM. Moreover, we propose hierarchical model aggregation as a strategy to reduce redundant token transmissions. At the edge server level, client updates are aggregated to refine local routing policies, while global coordination across clusters further synchronizes decision boundaries. This layered approach ensures that repeated uncertainty patterns are captured and resolved locally, significantly reducing unnecessary LLM queries. Extensive simulations on large-scale news classification tasks demonstrate that FedHLM achieves over 95% reduction in LLM transmissions with negligible accuracy loss, highlighting its potential for scalable and efficient edge-artificial intelligence (AI) deployment.

We consider the problem of privately estimating the mean of vectors distributed across different nodes of an unreliable wireless network, where communications between nodes can fail intermittently. We adopt a semi-decentralized setup, wherein to mitigate the impact of intermittently connected links, nodes can collaborate with their neighbors to compute a local consensus, which they relay to a central server. In such a setting, the communications between any pair of nodes must ensure that the privacy of the nodes is rigorously maintained to prevent unauthorized information leakage. We study the tradeoff between collaborative relaying and privacy leakage due to the data sharing among nodes and, subsequently, propose PRICER: Private Collaborative Estimation via Relaying, a differentially private collaborative algorithm for mean estimation to optimize this tradeoff. The privacy guarantees of PRICER arise (i) implicitly, by exploiting the inherent stochasticity of the flaky network connections, and (ii) explicitly, by adding Gaussian perturbations to the estimates exchanged by the nodes. Local and central privacy guarantees are provided against eavesdroppers who can observe different signals, such as the communications amongst nodes during local consensus and (possibly multiple) transmissions from the relays to the central server. We substantiate our theoretical findings with numerical simulations. Our implementation is available at https://github.com/rajarshisaha95/private-collaborative-relaying .