arXiv subjects

Xiang Cheng

Publications and source records attributed to Xiang Cheng.

At least 55 records · Page 3Linked to original sources

arXiv2026

Multi-Modal Sensing Residual-Corrected GNN for mmWave Path Loss Prediction via Synesthesia of Machines

To support sixth-generation (6G)-enabled intelligent transportation systems (ITSs), a multi-modal sensing residual-corrected graph neural network (MM-ResGNN) framework is proposed for millimeter-wave (mmWave) path loss prediction in vehicular communications for the first time. The propagation environment is formulated as an environment sensing path loss graph (ESPL-Graph), where nodes represent the transmitter (Tx) and receiver (Rx) entities and edges jointly describe Tx--Rx transmission links and Rx--Rx spatial correlation links. Meanwhile, a geometry-driven physical baseline is introduced to decouple deterministic attenuation trends from stochastic residual variations. A vehicular multi-modal path loss dataset (VMMPL) is constructed, which covers three representative scenarios, including the urban wide lane, urban crossroad, and suburban forking road environments, and achieves precise alignment between RGB images and global semantic information in the physical space, and link-level ray-tracing (RT)-based path loss data in the electromagnetic space. In MM-ResGNN, topology-aware graph representations and fine-grained visual semantics are synergistically integrated through a gated fusion mechanism to estimate the path loss residual relative to the physical baseline. Experimental results demonstrate that MM-ResGNN achieves significant improvements over empirical models and conventional data-driven baselines, with a normalized mean squared error (NMSE) of 0.0098, a mean absolute error (MAE) of 5.7991~dB, and a mean absolute percentage error (MAPE) of 5.0498\%. Furthermore, MM-ResGNN exhibits robust cross-scenario generalization through a few-shot fine-tuning strategy, enabling accurate path loss prediction in unseen vehicular environments with limited labeled data.

eess.SP↗

arXiv2026

LRT-Diffusion: Calibrated Risk-Aware Guidance for Diffusion Policies

Diffusion policies are competitive for offline reinforcement learning (RL) but are typically guided at sampling time by heuristics that lack a statistical notion of risk. We introduce LRT-Diffusion, a risk-aware sampling rule that treats each denoising step as a sequential hypothesis test between the unconditional prior and the state-conditional policy head. Concretely, we accumulate a log-likelihood ratio and gate the conditional mean with a logistic controller whose threshold tau is calibrated once under H0 to meet a user-specified Type-I level alpha. This turns guidance from a fixed push into an evidence-driven adjustment with a user-interpretable risk budget. Importantly, we deliberately leave training vanilla (two heads with standard epsilon-prediction) under the structure of DDPM. LRT guidance composes naturally with Q-gradients: critic-gradient updates can be taken at the unconditional mean, at the LRT-gated mean, or a blend, exposing a continuum from exploitation to conservatism. We standardize states and actions consistently at train and test time and report a state-conditional out-of-distribution (OOD) metric alongside return. On D4RL MuJoCo tasks, LRT-Diffusion improves the return-OOD trade-off over strong Q-guided baselines in our implementation while honoring the desired alpha. Theoretically, we establish level-alpha calibration, concise stability bounds, and a return comparison showing when LRT surpasses Q-guidance-especially when off-support errors dominate. Overall, LRT-Diffusion is a drop-in, inference-time method that adds principled, calibrated risk control to diffusion policies for offline RL.

cs.LG↗

arXiv2026

GKP-inspired high-dimensional superdense coding with energy-time entanglement

Superdense coding, the application of entanglement to boost classical communication capacity, is a cornerstone of quantum communication. In this paper, we propose a high-dimensional superdense coding protocol using energy-time entangled states. These states are biphoton frequency combs, an example of entangled time-frequency Gottesman-Kitaev-Preskill (TFGKP) states or time-frequency grid states. Inspired by GKP codes, our protocol involves discretizing the continuous time and frequency degrees of freedom and encoding information by time-frequency displacements. This approach leverages the inherently large Hilbert space found in quantum frequency combs, with resilience against both temporal and spectral errors. In addition to describing the theoretical structure of the protocol, we propose an experimental implementation using standard telecommunication components, time-resolving single-photon detectors and a frequency beamsplitter. We also analyze the effect of experimental noise and errors on the channel capacity of the protocol. We demonstrate that for realistic experimental parameters, contemporary technologies satisfy the prerequisites for superdense coding with biphoton frequency combs, achieving a transmission rate of approximately 8.91 bits per transmitted photon (equivalent to 481 distinguishable messages with asymptotically vanishing errors). This more than doubles the previously highest transmission rate of 4 bits achieved by the Kwiat-Weinfurter scheme, while also having competitive optical loss. Furthermore, our results beat the rate achievable using a single-photon frequency comb with identical parameters by 4.6 times. Our protocol thus represents an experimentally feasible application of time-frequency grid states to entanglement-assisted communication, contributing to the active fields of continuous-variable and high-dimensional quantum information.

quant-ph↗

arXiv2026

In-Context Semi-Supervised Learning

There has been significant recent interest in understanding the capacity of Transformers for in-context learning (ICL), yet most theory focuses on supervised settings with explicitly labeled pairs. In practice, Transformers often perform well even when labels are sparse or absent, suggesting crucial structure within unlabeled contextual demonstrations. We introduce and study in-context semi-supervised learning (IC-SSL), where a small set of labeled examples is accompanied by many unlabeled points, and show that Transformers can leverage the unlabeled context to learn a robust, context-dependent representation. This representation enables accurate predictions and markedly improves performance in low-label regimes, offering foundational insights into how Transformers exploit unlabeled context for representation learning within the ICL framework.

cs.LG↗

arXiv2026

Synesthesia of Machines (SoM)-Aided Online FDD Precoding via Heterogeneous Multi-Modal Sensing: A Vertical Federated Learning Approach

This paper investigates a heterogeneous multi-vehicle, multi-modal sensing (H-MVMM) aided online precoding problem. The proposed H-MVMM scheme utilizes a vertical federated learning (VFL) framework to minimize pilot sequence length and optimize the sum rate. This offers a promising solution for reducing latency in frequency division duplexing systems. To achieve this, three preprocessing modules are designed to transform raw sensory data into informative representations relevant to precoding. The approach effectively addresses local data heterogeneity arising from diverse on-board sensor configurations through a well-structured VFL training procedure. Additionally, a label-free online model updating strategy is introduced, enabling the H-MVMM scheme to adapt its weights flexibly. This strategy features a pseudo downlink channel state information label simulator (PCSI-Simulator), which is trained using a semi-supervised learning (SSL) approach alongside an online loss function. Numerical results show that the proposed method can closely approximate the performance of traditional optimization techniques with perfect channel state information, achieving a significant 90.6\% reduction in pilot sequence length.

eess.SP↗

arXiv2026

Inertia-Dilatancy Interplay Governs Shear-Thickening Drop Impact

Combining high-speed photography with direct force measurements, we investigate the impact dynamics of drops of cornstarch-water mixtures -- a premier example of shear-thickening fluids -- across a wide range of impact conditions. Our study identifies three distinct impact regimes. In addition to the liquid-like and solid-like behaviors generally expected for the impact-induced response of shear-thickening fluids, we uncover a counterintuitive regime in which high-concentration cornstarch-water mixtures display a liquid-like response at the onset of impact when shear rates are high and only transition to a solid-like behavior at later times as shear rates reduce. By integrating the classic drop-impact theory with the Reynolds-Darcy mechanism for dilatancy, we develop a unified model that quantitatively describes the impact dynamics of shear-thickening drops across all regimes. Our work reveals the unexpected response of shear-thickening fluids to ultra-fast deformation and advances fundamental understanding of drop impact for complex fluids.

physics.flu-dyn↗

arXiv2026

Tiny-WiFo: A Lightweight Wireless Foundation Model for Channel Prediction via Multi-Component Adaptive Knowledge Distillation

The massive scale of Wireless Foundation Models (FMs) hinders their real-time deployment on edge devices. This letter moves beyond standard knowledge distillation by introducing a novel Multi-Component Adaptive Knowledge Distillation (MCAKD) framework. Key innovations include a Cross-Attention-Based Knowledge Selection (CA-KS) module that selectively identifies critical features from the teacher model, and an Autonomous Learning-Passive Learning (AL-PL) strategy that balances knowledge transfer with independent learning to achieve high training efficiency at a manageable computational cost. When applied to the WiFo FM, the distilled Tiny-WiFo model, with only 5.5M parameters, achieves a 1.6 ms inference time while retaining over 98% of WiFo's performance and its crucial zero-shot generalization capability, making real-time FM deployment viable.

eess.SP↗

arXiv2026

Large Wireless Foundation Models: Stronger over Bigger

AI-communication integration is widely regarded as a core enabling technology for 6G. Most existing AI-based physical-layer designs rely on task-specific models that are separately tailored to individual modules, resulting in poor generalization. In contrast, communication systems are inherently general-purpose and should support broad applicability and robustness across diverse scenarios. Foundation models offer a promising solution through strong reasoning and generalization, yet wireless-system constraints hinder a direct transfer of large language model (LLM)-style success to the wireless domain. Therefore, we introduce the concept of large wireless foundation models (LWFMs) and present a novel framework for empowering the physical layer with foundation models under wireless constraints. Specifically, we propose two paradigms for realizing LWFMs, including leveraging existing general-purpose foundation models and building novel wireless foundation models. Based on recent progress, we distill two roadmaps for each paradigm and formulate design principles under wireless constraints. We further provide case studies of LWFM-empowered wireless systems to intuitively validate their advantages. Finally, we characterize the notion of "large" in LWFMs through a multidimensional analysis of existing work and outline promising directions for future research.

eess.SP↗

arXiv2026

GP-DHT: A Dual-Head Transformer with Contras-tive Learning for Predicting Gene Regulatory Rela-tionships across Species from Single-Cell Data

Gene regulatory networks (GRNs) are essential for understanding cell fate decisions and disease mechanisms, yet cross-species GRN inference from single-cell RNA-seq data remains challenging due to noise, sparsity, and cross-species distribution shifts. We propose GP-DHT (GenePair DualHeadTransformer), a cross-species single-cell GRN inference framework that models genes and cells in a heterogeneous graph with multi-level expression relations and learns structured regulatory representations via multi-relational graph attention. A dual-head Transformer further captures local gene pair regulatory dependencies and global cross-cell interaction patterns. To improve robustness under sparse and cross-species settings, GP-DHT introduces gene pair level supervised contrastive learning. Experiments on seven BEELINE benchmark datasets show consistent gains over representative baselines, improving AUROC and AUPRC by approximately 5 to 7 percent on most datasets. GP-DHT also recovers known regulatory modules and helps distinguish conserved from species-specific regulations.

q-bio.GN↗

arXiv2026

WiFo-E: A Scalable Wireless Foundation Model for End-to-End FDD Precoding in Communication Networks

Accurate precoding in massive multiple-input multiple-output (MIMO) frequency-division duplexing (FDD) systems relies on efficient channel state information (CSI) acquisition. End-to-end learning frameworks improve performance by jointly optimizing this process, but they lack scalability and fail to generalize across different system configurations, such as varying numbers of antennas and users. To overcome this limitation, we introduce WiFo-E, a wireless foundation model designed for scalable end-to-end precoding. WiFo-E employs multi-task pretraining on a diverse set of configurations to learn transferable representations of underlying wireless principles. Central to the model is a sparse Mixture-of-Experts (MoE) Transformer architecture, which mitigates task interference and enhances training efficiency by activating specialized parameter subsets adaptively. Extensive simulations demonstrate that WiFo-E outperforms conventional per-configuration training and shows strong generalization to unseen system configurations, providing a flexible and efficient foundation for adaptive massive MIMO precoding.

eess.SP↗

arXiv2026

Revisiting Chain-of-Thought Prompting: Zero-shot Can Be Stronger than Few-shot

In-Context Learning (ICL) is an essential emergent ability of Large Language Models (LLMs), and recent studies introduce Chain-of-Thought (CoT) to exemplars of ICL to enhance the reasoning capability, especially in mathematics tasks. However, given the continuous advancement of model capabilities, it remains unclear whether CoT exemplars still benefit recent, stronger models in such tasks. Through systematic experiments, we find that for recent strong models such as the Qwen2.5 series, adding traditional CoT exemplars does not improve reasoning performance compared to Zero-Shot CoT. Instead, their primary function is to align the output format with human expectations. We further investigate the effectiveness of enhanced CoT exemplars, constructed using answers from advanced models such as \texttt{Qwen2.5-Max} and \texttt{DeepSeek-R1}. Experimental results indicate that these enhanced exemplars still fail to improve the model's reasoning performance. Further analysis reveals that models tend to ignore the exemplars and focus primarily on the instructions, leading to no observable gain in reasoning ability. Overall, our findings highlight the limitations of the current ICL+CoT framework in mathematical reasoning, calling for a re-examination of the ICL paradigm and the definition of exemplars.

cs.CL↗

arXiv2026

AMAP Agentic Planning Technical Report

We present STAgent, an agentic large language model tailored for spatio-temporal understanding, designed to solve complex tasks such as constrained point-of-interest discovery and itinerary planning. STAgent is a specialized model capable of interacting with ten distinct tools within spatio-temporal scenarios, enabling it to explore, verify, and refine intermediate steps during complex reasoning. Notably, STAgent effectively preserves its general capabilities. We empower STAgent with these capabilities through three key contributions: (1) a stable tool environment that supports over ten domain-specific tools, enabling asynchronous rollout and training; (2) a hierarchical data curation framework that identifies high-quality data like a needle in a haystack, curating high-quality queries by retaining less than 1\% of the raw data, emphasizing both diversity and difficulty; and (3) a cascaded training recipe that starts with a seed SFT stage acting as a guardian to measure query difficulty, followed by a second SFT stage fine-tuned on queries with high certainty, and an ultimate RL stage that leverages data of low certainty. Initialized with Qwen3-30B-A3B to establish a strong SFT foundation and leverage insights into sample difficulty, STAgent yields promising performance on TravelBench while maintaining its general capabilities across a wide range of general benchmarks, thereby demonstrating the effectiveness of our proposed agentic model.

cs.AI↗

arXiv2026

WiFo-MUD: Wireless Foundation Model for Heterogeneous Multi-User Demodulator

Multi-user signal demodulation is critical to wireless communications, directly impacting transmission reliability and efficiency. However, existing demodulators underperform in generic multi-user environments: classical demodulators struggle to balance accuracy and complexity, while deep learning-based methods lack adaptability under heterogeneous configurations. Although diffusion models have been introduced for demodulation, their flexibility remains limited for practical use. To address these issues, this work proposes WiFo-MUD, a universal diffusion-based foundation model for multi-user demodulation. The model aligns inter-user signal-to-noise ratio imbalance and performs conditional denoising via a customized backbone. Furthermore, a communication-aware consistency distillation method and a dynamic user-grouping strategy are devised to enhance inference. WiFo-MUD achieves state-of-the-art results on large-scale heterogeneous datasets, demonstrating efficient inference and strong generalization across varying system configurations.

eess.SP↗

arXiv2025

NOC4SC: A Bandwidth-Efficient Multi-User Semantic Communication Framework for Interference-Resilient Transmission

With the explosive growth of connected devices and emerging applications, current wireless networks are encountering unprecedented demands for massive user access, where the inter-user interference has become a critical challenge to maintaining high quality of service (QoS) in multi-user communication systems. To tackle this issue, we propose a bandwidth-efficient semantic communication paradigm termed Non-Orthogonal Codewords for Semantic Communication (NOC4SC), which enables simultaneous same-frequency transmission without spectrum spreading. By leveraging the Swin Transformer, the proposed NOC4SC framework enables each user to independently extract semantic features through a unified encoder-decoder architecture with shared network parameters across all users, which ensures that the user's data remains protected from unauthorized decoding. Furthermore, we introduce an adaptive NOC and SNR Modulation (NSM) block, which employs deep learning to dynamically regulate SNR and generate approximately orthogonal semantic features within distinct feature subspaces, thereby effectively mitigating inter-user interference. Extensive experiments demonstrate the proposed NOC4SC achieves comparable performance to the DeepJSCC-PNOMA and outperforms other multi-user SemCom baseline methods.

eess.IV↗

arXiv2025

LLMs for Explainable Business Decision-Making: A Reinforcement Learning Fine-Tuning Approach

Artificial Intelligence (AI) models increasingly drive high-stakes consumer interactions, yet their decision logic often remains opaque. Prevailing explainable AI techniques rely on post hoc numerical feature attributions, which fail to provide coherent narratives behind model decisions. Large language models (LLMs) present an opportunity to generate natural-language explanations, but three design challenges remain unresolved: explanations must be both decision-correct and faithful to the factors that drive the prediction; they should be able to serve multiple audiences without shifting the underlying decision rule; and they should be trained in a label-efficient way that does not depend on large corpora of human-scored explanations. To address these challenges, we introduce LEXMA (LLM-based EXplanations for Multi-Audience decisions), a reinforcement-learning-based fine-tuning framework that produces narrative-driven, audience-appropriate explanations. LEXMA combines reflection-augmented supervised fine-tuning with two stages of Group Relative Policy Optimization (GRPO). Specifically, it fine-tunes two separate parameter sets to improve decision correctness and satisfy stylistic requirements for different audiences, using reward signals that do not rely on human-annotated explanations. We instantiate LEXMA in the context of mortgage approval decisions. Results demonstrate that LEXMA yields significant improvements in predictive performance compared with other LLM baselines. Moreover, human evaluations show that expert-facing explanations generated by our approach are more risk-focused, and consumer-facing explanations are clearer, more actionable, and more polite. Our study contributes a cost-efficient, systematic LLM fine-tuning approach to enhance explanation quality for business decisions, offering strong potential for scalable deployment of transparent AI systems.

cs.CL↗

arXiv2025

SITP: A High-Reliability Semantic Information Transport Protocol Without Retransmission for Semantic Communication

With the evolution of 6G networks, modern communication systems are facing unprecedented demands for high reliability and low latency. However, conventional transport protocols are designed for bit-level reliability, failing to meet the semantic robustness requirements. To address this limitation, this paper proposes a novel Semantic Information Transport Protocol (SITP), which achieves TCP-level reliability and UDP level latency by verifying only packet headers while retaining potentially corrupted payloads for semantic decoding. Building upon SITP, a cross-layer analytical model is established to quantify packet-loss probability across the physical, data-link, network, transport, and application layers. The model provides a unified probabilistic formulation linking signal noise rate (SNR) and packet-loss rate, offering theoretical foundation into end-to-end semantic transmission. Furthermore, a cross-image feature interleaving mechanism is developed to mitigate consecutive burst losses by redistributing semantic features across multiple correlated images, thereby enhancing robustness in burst-fade channels. Extensive experiments show that SITP offers lower latency than TCP with comparable reliability at low SNRs, while matching UDP-level latency and delivering superior reconstruction quality. In addition, the proposed cross-image semantic interleaving mechanism further demonstrates its effectiveness in mitigating degradation caused by bursty packet losses.

eess.IV↗

arXiv2025

Embodied Intelligent Wireless (EIW): Synesthesia of Machines Empowered Wireless Communications

The evolution toward the sixth-generation (6G) and beyond mobile communication systems is marked by a fundamental shift from merely connecting devices to enabling pervasive and embodied intelligence. While recent advances in artificial intelligence (AI)-native wireless communication designs have achieved remarkable progress, the prevailing paradigm remains limited to static, modular AI substitutions. This approach fails to meet the core requirements of future wireless networks: the ability to continuously perceive, adapt to, and interact with the dynamic wireless environment. To bridge this gap, this paper introduces Embodied Intelligent Wireless (EIW), a novel communication paradigm inspired by embodied intelligence, which redefines the communication node as an active, environment-aware, and evolving entity. EIW is built around an observation-decision-action paradigm, comprising: multi-dimensional observations for comprehensive awareness of the environment and system states, a unified decision module for orchestrating multiple wireless agents in an interpretable manner, and actions where wireless agents exert practical effects on both the environment and communication systems. Furthermore, two enabling technologies, wireless world models as well as self-update and self-evolution mechanisms, are introduced to support training efficiency improvement, counterfactual evaluation, and better adaptation. Unlike existing communication systems, EIW envisions future communication systems not as passive data pipelines, but as intelligent entities that continuously interact with and co-evolve alongside their environments. Finally, through simulations, we showcase the advantages of the proposed EIW paradigm and its enabling techniques in shaping the design of wireless communication nodes.

eess.SP↗

arXiv2025

TinyChemVL: Advancing Chemical Vision-Language Models via Efficient Visual Token Reduction and Complex Reaction Tasks

While Vision Language Models (VLMs) have demonstrated remarkable capabilities in general visual understanding, their application in the chemical domain has been limited, with previous works predominantly focusing on text and thus overlooking critical visual information, such as molecular structures. Current approaches that directly adopt standard VLMs for chemical tasks suffer from two primary issues: (i) computational inefficiency of processing entire chemical images with non-informative backgrounds. (ii) a narrow scope on molecular-level tasks that restricts progress in chemical reasoning. In this work, we propose \textbf{TinyChemVL}, an efficient and powerful chemical VLM that leverages visual token reduction and reaction-level tasks to improve model efficiency and reasoning capacity. Also, we propose \textbf{ChemRxn-V}, a reaction-level benchmark for assessing vision-based reaction recognition and prediction tasks. Directly predicting reaction products from molecular images poses a non-trivial challenge, as it requires models to integrate both recognition and reasoning capacities. Our results demonstrate that with only 4B parameters, TinyChemVL achieves superior performance on both molecular and reaction tasks while demonstrating faster inference and training speeds compared to existing models. Notably, TinyChemVL outperforms ChemVLM while utilizing only 1/16th of the visual tokens. This work builds efficient yet powerful VLMs for chemical domains by co-designing model architecture and task complexity.

cs.CV↗