Songyang Zhang

Radiomap serves as a vital tool for wireless network management and deployment by providing spatial knowledge of signal strength and coverage. However, increasingly complex radio propagation environments pose greater challenges for radiomap estimation (RME). This work proposes a knowledge-guided RME framework that integrates Kolmogorov–Arnold Networks (KAN) with Mixture-of-Experts (MoE), named RadioKMoE. Specifically, our KAN module predicts an initial coarse coverage map by leveraging its strength to approximate physics models and global radio propagation patterns. The initial map, together with environmental information, drives our MoE network for precise radiomap estimation. Unlike conventional deep learning models, the MoE module features expert networks specializing in distinct radiomap patterns to improve local details while preserving global consistency. Experimental results in both multi- and single-band RME demonstrate enhanced accuracy and robustness of the proposed RadioKMoE in radiomap estimation.

The recent empirical success of Mamba and other selective state space models (SSMs) has renewed interest in non-attention architectures for sequence modeling, yet their theoretical foundations remain underexplored. We present a first-step analysis of generalization and learning dynamics for a simplified but representative Mamba block: a single-layer, single-head selective SSM with input-dependent gating, followed by a two-layer MLP trained via gradient descent (GD). Our study adopts a structured data model with tokens that include both class-relevant and class-irrelevant patterns under token-level noise and examines two canonical regimes: majority-voting and locality-structured data sequences. We prove that the model achieves guaranteed generalization by establishing non-asymptotic sample complexity and convergence rate bounds, which improve as the effective signal increases and the noise decreases. Furthermore, we show that the gating vector aligns with class-relevant features while ignoring irrelevant ones, thereby formalizing a feature-selection role similar to attention but realized through selective recurrence. Numerical experiments on synthetic data justify our theoretical results. Overall, our results provide principled insight into when and why Mamba-style selective SSMs learn efficiently, offering a theoretical counterpoint to Transformer-centric explanations.

Contrastive learning has emerged as a powerful framework for learning generalizable representations, yet its theoretical understanding remains limited, particularly under imbalanced data distributions that are prevalent in real-world applications. Such an imbalance can degrade representation quality and induce biased model behavior, yet a rigorous characterization of these effects is lacking. In this work, we develop a theoretical framework to analyze the training dynamics of contrastive learning with Transformer-based encoders under imbalanced data. Our results reveal that neuron weights evolve through three distinct stages of training, with different dynamics for majority features, minority features, and noise. We further show that minority features reduce representational capacity, increase the need for more complex architectures, and hinder the separation of ground-truth features from noise. Inspired by these neuron-level behaviors, we show that pruning restores performance degraded by imbalance and enhances feature separation, offering both conceptual insights and practical guidance. Major theoretical findings are validated through numerical experiments.

Recent advances in generative artificial intelligence (AI) have led to rising interest in federated learning (FL) based on generative adversarial network (GAN) models. GAN-based FL shows promises in many communication and network applications, such as edge computing and the Internet of Things. In the context of FL, GANs can capture the underlying client data structure, and regenerate samples resembling the original data distribution without compromising data privacy. Although most existing GAN-based FL works focus on training a global model, some scenarios exist where personalized FL (PFL) can be more desirable when incorporating client data heterogeneity in terms of distinct data distributions, feature spaces, and labels. To cope with client heterogeneity in GAN-based FL, we propose a novel GAN sharing and aggregation strategy for PFL that can efficiently characterize client heterogeneity in different settings. More specifically, our proposed PFL-GAN first learns the similarities among clients before implementing a weighted collaborative data aggregation. Our empirical results through rigorous experimentation on several well-known datasets demonstrate the effectiveness of PFL-GAN.

The recent rise of semantic-style communications has fostered the development of goal-oriented communications (GO-COMs), facilitating remarkably efficient multimedia information transmissions. The concept of GO-COMs leverages advanced artificial intelligence (AI) tools to address the rising demand for bandwidth efficiency in applications, such as edge computing and the Internet of Things (IoT). Unlike traditional communication systems focusing on source data accuracy, GO-COMs provide intelligent message delivery catering to the special needs critical to accomplishing downstream tasks at the receiver. In this work, we present a novel GO-COM framework, namely LaMI-GO, that utilizes emerging generative AI for better quality of service (QoS) with ultrahigh communication efficiency. Specifically, we design our LaMI-GO system backbone based on a latent diffusion model followed by a vector-quantized generative adversarial network (VQGAN) for efficient latent embedding and information representation. The system trains a common-feature codebook for the receiver side. Our experimental results demonstrate substantial improvement in perceptual quality, accuracy of downstream tasks, and bandwidth consumption over the state-of-the-art GO-COM systems and establish the power of our proposed LaMI-GO communication framework.

The rapid development of artificial intelligence has driven smart health with next-generation wireless communication technologies, stimulating exciting applications in remote diagnosis and intervention. To enable a timely and effective response for remote healthcare, efficient transmission of medical data through noisy channels with limited bandwidth emerges as a critical challenge. In this work, we propose a novel diffusion-based semantic communication framework, namely DiSC-Med, for the medical image transmission, where medical-enhanced compression and denoising blocks are developed for bandwidth efficiency and robustness, respectively. Unlike conventional pixel-wise communication framework, our proposed DiSC-Med is able to capture the key semantic information and achieve superior reconstruction performance with ultra-high bandwidth efficiency against noisy channels. Extensive experiments on real-world medical datasets validate the effectiveness of our framework, demonstrating its potential for robust and efficient telehealth applications.

The integration of Federated Learning (FL) and Mixture-of-Experts (MoE) presents a compelling pathway for training more powerful, large-scale artificial intelligence models (LAMs) on decentralized data while preserving privacy. However, efficient federated training of these complex MoE-structured LAMs is hindered by significant system-level challenges, particularly in managing the interplay between heterogeneous client resources and the sophisticated coordination required for numerous specialized experts. This article highlights a critical, yet underexplored concept: the absence of robust quantitative strategies for dynamic client-expert alignment that holistically considers varying client capacities and the imperative for system-wise load balancing. Specifically, we propose a conceptual system design for intelligent client-expert alignment that incorporates dynamic fitness scoring, global expert load monitoring, and client capacity profiling. By tackling these systemic issues, we can unlock more scalable, efficient, and robust training mechanisms {with fewer communication rounds for convergence}, paving the way for the widespread deployment of large-scale federated MoE-structured LAMs in edge computing with ultra-high communication efficiency.

Radiomap characterizes geographical radio spectrum coverage and can facilitate resource allocation and management of wireless networks. One practical radiomap estimation (RME) task is to form a full radiomap from sparse samples collected by sensors or mobile devices. Often, traditional RME approaches focus on statistical data distributions without exploiting the underlying spatial correlations among sparse observations. Utilizing geometric/geographical path correlation, this work proposes a novel dual-phase RME method based on graph neural networks. In this Dual-phase Graph-based Radiomap Estimation (Dual-GRE) framework, the first phase integrates graph attention (GAT) networks with radio propagation models to construct a coarse-resolution (CR) radiomap to embed the spatial information and physical principles. Phase 2 utilizes a deep convolution neural network that uses the CR radiomap and landscape information to derive fine-resolution radiomaps. Our experimental results demonstrate the power of physics-integrated GAT in capturing the spatial spectrum information, together with the efficiency of the proposed Dual-GRE in radiomap estimation.