Seong-Lyun Kim

To support emerging language-based applications using dispersed and heterogeneous computing resources, the hybrid language model (HLM) offers a promising architecture, where an on-device small language model (SLM) generates draft tokens that are validated and corrected by a remote large language model (LLM). However, the original HLM suffers from substantial communication overhead, as the LLM requires the SLM to upload the full vocabulary distribution for each token. Moreover, both communication and computation resources are wasted when the LLM validates tokens that are highly likely to be accepted. To overcome these limitations, we propose communication-efficient and uncertainty-aware HLM (CU-HLM). In CU-HLM, the SLM transmits truncated vocabulary distributions only when its output uncertainty is high. We validate the feasibility of this opportunistic transmission by discovering a strong correlation between SLM’s uncertainty and LLM’s rejection probability. Furthermore, we theoretically derive optimal uncertainty thresholds and optimal vocabulary truncation strategies. Simulation results show that, compared to standard HLM, CU-HLM achieves up to 206× higher token throughput by skipping 74.8% transmissions with 97.4% vocabulary compression, while maintaining 97.4% accuracy.

This letter introduces a communication-efficient multi-modal semantic communication (MMSC) framework over wireless networks, termed Residual Hierarchical-Adaptive Modality Control (ReHi-AMC). Under an outage-based transmission without CSI feedback, ReHi-AMC preserves generative reconstruction quality. The framework assigns cross-modal transmission priorities and progressively encodes spatial sketches through a hierarchical multi-resolution representation. To further enhance payload efficiency, residual information between hierarchical levels is compressed via sparse residual encoding, reducing redundancy while maintaining essential structural cues. Numerical evaluations on a Stable Diffusion–based generative architecture demonstrate up to 46.47% improvement in CLIP score, with particularly pronounced gains in low-SNR regimes, confirming robustness against channel outages compared to single-scale baselines without residual encoding.

The importance of Radio Access Network (RAN) in support Artificial Intelligence (AI) application services has grown significantly, underscoring the need for an integrated approach that considers not only network efficiency but also AI performance. In this paper we focus on a semantic generative communication (SGC) framework for image inpainting application. Specifically, the transmitter sends semantic information, i.e., semantic masks and textual descriptions, while the receiver utilizes a conditional diffusion model on a base image, using them as conditioning data to produce the intended image. In this framework, we propose a bandwidth allocation scheme designed to maximize bandwidth efficiency while ensuring generation performance. This approach is based on our finding of a Semantic Deadline--the minimum time that conditioning data is required to be injected to meet a given performance threshold--within the multi-modal SGC framework. Given this observation, the proposed scheme allocates limited bandwidth so that each semantic information can be transmitted within the corresponding semantic deadline. Experimental results corroborate that the proposed bandwidth allocation scheme achieves higher generation performance in terms of PSNR for a given bandwidth compared to traditional schemes that do not account for semantic deadlines.

Recent advances in semantic communication (SC) have introduced neural network (NN)-based transceivers that convey semantic representation (SR) of signals such as images. However, these NNs are trained over diverse image distributions and thus often fail to reconstruct fine-grained image-specific details. To overcome this limited reconstruction fidelity, we propose an extended SC framework, hybrid semantic communication (HSC), which supplements SR with complementary representation (CR) capturing residual image-specific information. The CR is constructed at the transmitter, and is combined with the actual SC outcome at the receiver to yield a high-fidelity recomposed image. While the transmission load of SR is fixed due to its NN-based structure, the load of CR can be flexibly adjusted to achieve a desirable fidelity. This controllability directly influences the final reconstruction error, for which we derive a closed-form expression and the corresponding optimal CR. Simulation results demonstrate that HSC substantially reduces MSE compared to the baseline SC without CR transmission across various channels and NN architectures.

The forthcoming 6G systems are expected to address a wide range of non-stationary tasks. This poses challenges to traditional medium access control (MAC) protocols that are static and predefined. In response, data-driven MAC protocols have recently emerged, offering ability to tailor their signaling messages for specific tasks. This article presents a novel categorization of these data-driven MAC protocols into three levels: Level 1 MAC. task-oriented neural protocols constructed using multi-agent deep reinforcement learning (MADRL); Level 2 MAC. neural network-oriented symbolic protocols developed by converting Level 1 MAC outputs into explicit symbols; and Level 3 MAC. language-oriented semantic protocols harnessing large language models (LLMs) and generative models. With this categorization, we aim to explore the opportunities and challenges of each level by delving into their foundational techniques. Drawing from information theory and associated principles as well as selected case studies, this study provides insights into the trajectory of data-driven MAC protocols and sheds light on future research directions.

Due to the emergence of various wireless sensing technologies, numerous positioning algorithms have been introduced in the literature, categorized into geometry-driven positioning (GP) and data-driven positioning (DP). These approaches have respective limitations, e.g., a non-line-of-sight issue for GP and the lack of a high-dimensional and labeled dataset for DP, which could be complemented by integrating both methods. To this end, this paper aims to introduce a novel principle called combinatorial data augmentation (CDA), a catalyst for the two approaches’ seamless integration. Specifically, GP-based data samples augmented from different positioning element combinations are called preliminary estimated locations (PELs), which can be used as high-dimensional inputs for DP. We confirm the CDA’s effectiveness from field experiments based on WiFi round-trip times (RTTs) and inertial measurement units (IMUs) by designing several CDA-based positioning algorithms. First, we show that CDA offers various metrics quantifying each PEL’s reliability, thereby extracting important PELs for WiFi RTT positioning. Second, CDA helps compute the observation error covariance matrix of a Kalman filter for fusing two position estimates derived by WiFi RTTs and IMUs. Third, we use the important PELs and the above position estimate as the corresponding input feature and the real-time label for fingerprint-based positioning as a representative DP algorithm. It provides accurate and reliable positioning results, with an average positioning error of 1.58 (m) and a standard deviation of 0.90 (m).

A distributed control of vehicle platooning is referred to as distributed consensus (DC) since many autonomous vehicles (AVs) reach a consensus to achieve coordinated movement with the same velocity and inter-distance. For DC control to be stable, each AV utilizes other AVs’ real-time position information obtained via vehicle-to-vehicle (V2V) communications. On the other hand, too many V2V links should be simultaneously established and frequently retrained, causing a longer communication latency due to frequent packet losses and thereby hampering stable DC. This paper proposes a novel DC algorithm called over-the-air consensus (AirCons), a joint communication-and-control design with two key features to overcome the above limitations. First, exploiting a wireless signal’s superposition and broadcasting properties renders every AV’s signal converge to a specific value. We show that the consensus value is proportional to the weighted average of participating AVs’ real-time positions and has a tight lower bound as the ground-truth average. In other words, the average position location can be directly estimated without the neighbor AVs’ positions, thereby achieving ultra-low latency data sharing. Next, the estimated average position is inputted into each AV’s controller to adjust its dynamics distributively. The average position, considered a real-time value due to its low latency, contributes to achieving the stability of vehicle platooning. We design AirCons based on New Radio architecture with its feasibility study by analyzing required radio resources, i.e., time and bandwidth. Through analytic and numerical studies, the effectiveness of the proposed AirCons is verified by showing a 16.40% control gain compared to the benchmark without the average position.

In computer vision, the vision transformer (ViT) has increasingly superseded the convolutional neural network (CNN) for improved accuracy and robustness. However, ViT's large model sizes and high sample complexity make it difficult to train on resource-constrained edge devices. Split learning (SL) emerges as a viable solution, leveraging server-side resources to train ViTs while utilizing private data from distributed devices. However, SL requires additional information exchange for weight updates between the device and the server, which can be exposed to various attacks on private training data. To mitigate the risk of data breaches in classification tasks, inspired from the CutMix regularization, we propose a novel privacy-preserving SL framework that injects Gaussian noise into smashed data and mixes randomly chosen patches of smashed data across clients, coined DP-CutMixSL. Our analysis demonstrates that DP-CutMixSL is a differentially private (DP) mechanism that strengthens privacy protection against membership inference attacks during forward propagation. Through simulations, we show that DP-CutMixSL improves privacy protection against membership inference attacks, reconstruction attacks, and label inference attacks, while also improving accuracy compared to DP-SL and DP-MixSL.