arXiv subjects

Xiang Cheng

Publications and source records attributed to Xiang Cheng.

At least 91 records · Page 5Linked to original sources

arXiv2025

Probabilistic Uncertain Reward Model

Reinforcement learning from human feedback (RLHF) is a critical technique for training large language models. However, conventional reward models based on the Bradley-Terry model (BTRM) often suffer from overconfidence when faced with inconsistent labels or out-of-distribution samples, leading to reward hacking, where the policy model blindly optimizes for proxy rewards while degrading true performance. This paper proposes the Probabilistic Uncertain Reward Model (PURM), which generalizes the Bradley-Terry model to learn the reward distributions that emerged from the preference data. We theoretically derive the loss function of PURM and introduce a novel method that uses the overlap between distributions to quantify uncertainty. Empirical results show that PURM outperforms existing methods with more accurate reward and sound uncertainty estimations, and sustains effective learning for more optimization steps and obtain higher maximum win rate in RLHF. The data and code of this paper are released at https://anonymous.4open.science/r/Probabilistic-Uncertain-Reward-Model/

cs.LG↗

arXiv2025

Reducing classical communication costs in multiplexed quantum repeaters using hardware-aware quasi-local policies

Future quantum networks will have nodes equipped with multiple quantum memories, allowing for multiplexing and entanglement distillation strategies in order to increase fidelities and reduce waiting times for end-to-end entanglement distribution. In this work, we introduce \textit{quasi-local} policies for multiplexed quantum repeater chains. In fully-local policies, nodes make decisions based only on knowledge of their own states. In our quasi-local policies, nodes have increased knowledge of the state of the repeater chain, but not necessarily full, global knowledge. Our policies exploit the observation that for most decisions the nodes have to make, they only need to have information about the connected region of the chain they belong to, and not the entire chain. In this way, we not only obtain improved performance over local policies, but we reduce the classical communication (CC) costs inherent to global-knowledge policies. Our policies also outperform the well-known and widely studied nested purification and doubling swapping policy in practically relevant parameter regimes. We also carefully examine the role of entanglement distillation. Via analytical and numerical results, we identify the parameter regimes in which distillation makes sense and is useful. In these regimes, we also address the question: "Should we distill before swapping, or vice versa?" Finally, to provide further practical guidance, we propose an experimental implementation of a multiplexing-based repeater chain, and experimentally demonstrate the key element, a high-dimensional biphoton frequency comb. We then evaluate the anticipated performance of our multiplexing-based policies in such a real-world network through simulation results for two concrete memory platforms, namely rare-earth ions and diamond vacancies.

quant-ph↗

arXiv2025

Dynamics of rotating helices in viscous fluid

We investigate the dynamics of a pair of rigid rotating helices in a viscous fluid, as a model for bacterial flagellar bundle and a prototype of microfluidic pumps. Combining experiments with hydrodynamic modeling, we examine how spacing and phase difference between the two helices affect their torque, flow field and fluid transport capacity at low Reynolds numbers. Hydrodynamic coupling reduces the torque when the helices rotate in phase at constant angular speed, but increases the torque when they rotate out of phase. We identify a critical phase difference, at which the hydrodynamic coupling vanishes despite the close spacing between the helices. A simple model, based on the flow characteristics and positioning of a single helix, is constructed, which quantitatively predicts the torque of the helical pair in both unbounded and confined systems. Lastly, we show the influence of spacing and phase difference on the axial flux and the pump efficiency of the helices. Our findings shed light on the function of bacterial flagella and provide design principles for efficient low-Reynolds-number pumps.

physics.flu-dyn↗

arXiv2025

On Understanding Attention-Based In-Context Learning for Categorical Data

In-context learning based on attention models is examined for data with categorical outcomes, with inference in such models viewed from the perspective of functional gradient descent (GD). We develop a network composed of attention blocks, with each block employing a self-attention layer followed by a cross-attention layer, with associated skip connections. This model can exactly perform multi-step functional GD inference for in-context inference with categorical observations. We perform a theoretical analysis of this setup, generalizing many prior assumptions in this line of work, including the class of attention mechanisms for which it is appropriate. We demonstrate the framework empirically on synthetic data, image classification and language generation.

stat.ML↗

arXiv2025

HyboWaveNet: Hyperbolic Graph Neural Networks with Multi-Scale Wavelet Transform for Protein-Protein Interaction Prediction

Protein-protein interactions (PPIs) are fundamental for deciphering cellular functions,disease pathways,and drug discovery.Although existing neural networks and machine learning methods have achieved high accuracy in PPI prediction,their black-box nature leads to a lack of causal interpretation of the prediction results and difficulty in capturing hierarchical geometries and multi-scale dynamic interaction patterns among proteins.To address these challenges, we propose HyboWaveNet,a novel deep learning framework that collaborates with hyperbolic graphical neural networks (HGNNs) and multiscale graphical wavelet transform for robust PPI prediction. Mapping protein features to Lorentz space simulates hierarchical topological relationships among biomolecules via a hyperbolic distance metric,enabling node feature representations that better fit biological a priori.HyboWaveNet inherently simulates hierarchical and scale-free biological relationships, while the integration of wavelet transforms enables adaptive extraction of local and global interaction features across different resolutions. Our framework generates node feature representations via a graph neural network under the Lorenz model and generates pairs of positive samples under multiple different views for comparative learning, followed by further feature extraction via multi-scale graph wavelet transforms to predict potential PPIs. Experiments on public datasets show that HyboWaveNet improves over both existing state-of-the-art methods. We also demonstrate through ablation experimental studies that the multi-scale graph wavelet transform module improves the predictive performance and generalization ability of HyboWaveNet. This work links geometric deep learning and signal processing to advance PPI prediction, providing a principled approach for analyzing complex biological systems

cs.LG↗

arXiv2025

Heterogeneous network drug-target interaction prediction model based on graph wavelet transform and multi-level contrastive learning

Drug-target interaction (DTI) prediction is a core task in drug development and precision medicine in the biomedical field. However, traditional machine learning methods generally have the black box problem, which makes it difficult to reveal the deep correlation between the model decision mechanism and the interaction pattern between biological molecules. This study proposes a heterogeneous network drug target interaction prediction framework, integrating graph neural network and multi scale signal processing technology to construct a model with both efficient prediction and multi level interpretability. Its technical breakthroughs are mainly reflected in the following three dimensions:Local global feature collaborative perception module. Based on heterogeneous graph convolutional neural network (HGCN), a multi order neighbor aggregation strategy is designed.Multi scale graph signal decomposition and biological interpretation module. A deep hierarchical node feature transform (GWT) architecture is proposed.Contrastive learning combining multi dimensional perspectives and hierarchical representations. By comparing the learning models, the node representations from the two perspectives of HGCN and GWT are aligned and fused, so that the model can integrate multi dimensional information and improve the prediction robustness. Experimental results show that our framework shows excellent prediction performance on all datasets. This study provides a complete solution for drug target discovery from black box prediction to mechanism decoding, and its methodology has important reference value for modeling complex biomolecular interaction systems.

q-bio.QM↗

arXiv2025

SynthSoM: A synthetic intelligent multi-modal sensing-communication dataset for Synesthesia of Machines (SoM)

Given the importance of datasets for sensing-communication integration research, a novel simulation platform for constructing communication and multi-modal sensory dataset is developed. The developed platform integrates three high-precision software, i.e., AirSim, WaveFarer, and Wireless InSite, and further achieves in-depth integration and precise alignment of them. Based on the developed platform, a new synthetic intelligent multi-modal sensing-communication dataset for Synesthesia of Machines (SoM), named SynthSoM, is proposed. The SynthSoM dataset contains various air-ground multi-link cooperative scenarios with comprehensive conditions, including multiple weather conditions, times of the day, intelligent agent densities, frequency bands, and antenna types. The SynthSoM dataset encompasses multiple data modalities, including radio-frequency (RF) channel large-scale and small-scale fading data, RF millimeter wave (mmWave) radar sensory data, and non-RF sensory data, e.g., RGB images, depth maps, and light detection and ranging (LiDAR) point clouds. The quality of SynthSoM dataset is validated via statistics-based qualitative inspection and evaluation metrics through machine learning (ML) via real-world measurements. The SynthSoM dataset is open-sourced and provides consistent data for cross-comparing SoM-related algorithms.

eess.SP↗

arXiv2025

Superconductivity and quantized anomalous Hall in rhombohedral graphene

Inducing superconducting correlations in chiral edge states is predicted to generate topologically protected zero energy modes with exotic quantum statistics. Experimental efforts to date have focused on engineering interfaces between superconducting materials typically amorphous metals and semiconducting quantum Hall or quantum anomalous Hall (QAH) systems. However, the interfacial disorder inherent in this approach can prevent the formation of isolated topological modes. An appealing alternative is to use low-density flat band materials where the ground state can be tuned between intrinsic superconducting and quantum anomalous Hall states using only the electric field effect. However, quantized transport and superconductivity have not been simultaneously achieved. Here, we show that rhombohedral tetralayer graphene aligned to a hexagonal boron nitride substrate hosts a quantized anomalous Hall state at superlattice filling $ν=-1$ as well as a superconducting state at $ν-3.5$ at zero magnetic field. Remarkably, gate voltage can also be used to actuate nonvolatile switching of the chirality in the quantum anomalous Hall state, allowing, in principle, arbitrarily reconfigurable networks of topological edge modes in locally gated devices. Thermodynamic compressibility measurements further reveal a topologically ordered fractional Chern insulator at $ν=2/3$-also stable at zero magnetic field-enabling proximity coupling between superconductivity and fractionally charged edge modes. Finally, we show that, as in rhombohedral bi- and trilayers, integrating a transition metal dichalcogenide layer to the heterostructure nucleates a new superconducting pocket, while leaving the topology of the $ν=-1$ quantum anomalous Hall state intact. Our results pave the way for a new generation of hybrid interfaces between superconductors and topological edge states in the low-disorder limit.

cond-mat.mes-hall↗

arXiv2025

Maximum spreading of impacting shear-thinning and shear-thickening drops

The maximum spreading of an impacting liquid drop is a key metric for characterizing the fundamental fluid process of drop impact. While extensively studied for Newtonian liquids, how far a non-Newtonian drop can spread upon impacting a solid substrate remains an open question. Here, by combining simulations, experiments, and scaling analyses, we establish a general framework for predicting the maximum spreading of drops of generalized Newtonian liquids, encompassing both shear-thinning and shear-thickening behaviors. Through an analysis of the energy budget at maximum spreading, we identify a characteristic shear rate that governs the viscous dissipation during drop impact. The finding allows us to map the spreading of non-Newtonian drops onto that of Newtonian drops, revealing the quantitative dependence of the maximum spreading diameter on various impact parameters and rheological properties of liquids. Our study addresses the long-standing challenge of understanding the impact dynamics of non-Newtonian drops, and provides valuable guidance for designing non-Newtonian liquids to achieve desired impact outcomes.

physics.flu-dyn↗

arXiv2025

Recent advances in high-dimensional mode-locked quantum frequency combs

High-dimensional entanglement in qudit states offers a promising pathway towards the realization of practical, large-scale quantum systems that are highly controllable. These systems can be leveraged for various applications, including advanced quantum information processing, secure communications, computation, and metrology. In this context, quantum frequency combs have a crucial role as they inherently support multiple modes in both temporal and frequency domains, while preserving a single spatial mode. The multiple temporal and frequency modes of quantum frequency combs facilitate the generation, characterization, and control of high-dimensional time-frequency entanglement in extensive quantum systems. In this review article, we provide an overview of recent technological advancements in high-dimensional energy-time entangled quantum frequency combs. We explore how these time-frequency qudits, achieved using scalable telecommunications-wavelength components, can empower the creation of large-scale quantum states. Advances in quantum frequency combs can unlock new capabilities and versatility for promising developments in quantum science and technology.

quant-ph↗

arXiv2025

WiFo: Wireless Foundation Model for Channel Prediction

Channel prediction permits to acquire channel state information (CSI) without signaling overhead. However, almost all existing channel prediction methods necessitate the deployment of a dedicated model to accommodate a specific configuration. Leveraging the powerful modeling and multi-task learning capabilities of foundation models, we propose the first space-time-frequency (STF) wireless foundation model (WiFo) to address time-frequency channel prediction tasks in a one-for-all manner. Specifically, WiFo is initially pre-trained over massive and extensive diverse CSI datasets. Then, the model will be instantly used for channel prediction under various CSI configurations without any fine-tuning. We propose a masked autoencoder (MAE)-based network structure for WiFo to handle heterogeneous STF CSI data, and design several mask reconstruction tasks for self-supervised pre-training to capture the inherent 3D variations of CSI. To fully unleash its predictive power, we build a large-scale heterogeneous simulated CSI dataset consisting of 160K CSI samples for pre-training. Simulations validate its superior unified learning performance across multiple datasets and demonstrate its state-of-the-art (SOTA) zero-shot generalization performance via comparisons with other full-shot baselines.

eess.SP↗

arXiv2025

Synesthesia of Machines (SoM)-Enhanced Wideband Multi-User CSI Learning With LiDAR Sensing

Light detection and ranging (LiDAR) has been utilized for optimizing wireless communications due to its ability to detect the environment. This paper explores the use of LiDAR in channel estimation for wideband multi-user multiple-input-multiple-output orthogonal frequency division multiplexing systems and introduces a LiDAR-enhanced Channel State Information (CSI) learning network (LE-CLN). By utilizing user positioning information, LE-CLN first calculates user-localized over-complete angular measurements. It then investigates the correlation between LiDAR and CSI, transforming raw LiDAR data into a low-complexity format embedded with signal propagation characteristics. LE-CLN also adapts the use of LiDAR based on channel conditions through attention mechanisms. Thanks to the unique wireless features offered by LiDAR, LE-CLN achieves higher estimation accuracy and spectrum efficiency compared to benchmarks, particularly in latency-sensitive applications where pilot transmissions are expected to be reduced.

eess.SP↗

arXiv2025

Synesthesia of Machines (SoM)-Aided FDD Precoding with Sensing Heterogeneity: A Vertical Federated Learning Approach

High complexity in precoding design for frequency division duplex systems necessitates streamlined solutions. Guided by Synesthesia of Machines (SoM), this paper introduces a heterogeneous multi-vehicle, multi-modal sensing aided precoding scheme within a vertical federated learning (VFL) framework, which significantly minimizes pilot sequence length while optimizing the system's sum rate. We address the challenges posed by local data heterogeneity due to varying on-board sensor configurations through a meticulously designed VFL training procedure. To extract valuable channel features from multi-modal sensing, we employ three distinct data preprocessing methods that convert raw data into informative representations relevant for precoding. Additionally, we propose an online training strategy based on VFL framework, enabling the scheme to adapt dynamically to fluctuations in user numbers. Numerical results indicate that our approach, utilizing short pilot sequences, closely approximates the performance of traditional optimization methods with perfect channel state information.

eess.SP↗

arXiv2025

Synesthesia of Machine (SoM)-Driven Analog Precoder Optimization for Enhanced ISAC Performance in Sub-THz Systems

Integrated sensing and communication (ISAC) is anticipated to be widely used in future sub-terahertz (sub-THz) systems. With the line-of-sight (LoS) propagation characteristics of sub-THz channels, ISAC transmitter design largely parallels analog precoder optimization. However, balancing both sensing and communication functionalities is challenging due to the beam squint effect in sub-THz systems, limiting ISAC performance gains. To overcome this, the unique design flexibility of sub-THz analog hardware is explored to better adapt to the electromagnetic characteristics of sub-THz channels. It is demonstrated that adjusting the equivalent channel through the analog precoder enhances dual-functional gains. Based on this, a near-optimal benchmark for analog precoder optimization is proposed. To address excessive algorithmic complexity, inspiration is drawn from the synesthesia of machine (SoM) to develop a lightweight complex-valued squint-aware network (CSP-Net). This network reduces complexity by utilizing both communication and sensing channel data, with an architecture tailored to specific data and task characteristics. The effectiveness of the proposed schemes is validated through simulations.

eess.SP↗

arXiv2025

Graph Transformers Dream of Electric Flow

We show theoretically and empirically that the linear Transformer, when applied to graph data, can implement algorithms that solve canonical problems such as electric flow and eigenvector decomposition. The Transformer has access to information on the input graph only via the graph's incidence matrix. We present explicit weight configurations for implementing each algorithm, and we bound the constructed Transformers' errors by the errors of the underlying algorithms. Our theoretical findings are corroborated by experiments on synthetic data. Additionally, on a real-world molecular regression task, we observe that the linear Transformer is capable of learning a more effective positional encoding than the default one based on Laplacian eigenvectors. Our work is an initial step towards elucidating the inner-workings of the Transformer for graph data. Code is available at https://github.com/chengxiang/LinearGraphTransformer

cs.LG↗

arXiv2025

Beam Pattern Modulation Embedded Hybrid Transceiver Optimization for Integrated Sensing and Communication

Integrated sensing and communication (ISAC) emerges as a promising technology for B5G/6G, particularly in the millimeter-wave (mmWave) band. However, the widely utilized hybrid architecture in mmWave systems compromises multiplexing gain due to the constraints of limited radio frequency chains. Moreover, additional sensing functionalities exacerbate the impairment of spectrum efficiency (SE). In this paper, we present an optimized beam pattern modulation-embedded ISAC (BPM-ISAC) transceiver design, which spares one RF chain for sensing and the others for communication. To compensate for the reduced SE, index modulation across communication beams is applied. We formulate an optimization problem aimed at minimizing the mean squared error (MSE) of the sensing beampattern, subject to a symbol MSE constraint. This problem is then solved by sequentially optimizing the analog and digital parts. Both the multi-aperture structure (MAS) and the multi-beam structure (MBS) are considered for the design of the analog part. We conduct theoretical analysis on the asymptotic pairwise error probability (APEP) and the Cramér-Rao bound (CRB) of direction of arrival (DoA) estimation. Numerical simulations validate the overall enhanced ISAC performance over existing alternatives.

eess.SP↗

arXiv2025

Microscopic contact line dynamics dictate the emergent behaviors of particle rafts

Fluid-fluid interfaces laden with discrete particles behave curiously like continuous elastic sheets, leading to their applications in emulsion and foam stabilization. Although existing continuum models can qualitatively capture the elastic buckling of these particle-laden interfaces -- often referred to as particle rafts -- under compression, they fail to link their macroscopic collective properties to the microscopic behaviors of individual particles. Thus, phenomena such as particle expulsion from the compressed rafts remain unexplained. Here, by combining systematic experiments with first-principle modeling, we reveal how the macroscopic mechanical properties of particle rafts emerge from particle-scale interactions. We construct a phase diagram that delineates the conditions under which a particle raft collapses via collective folding versus single-particle expulsion. Guided by this theoretical framework, we demonstrate control over the raft's failure mode by tuning the physicochemical properties of individual particles. Our study highlights the previously overlooked dual nature of particle rafts and exemplifies how collective dynamics can arise from discrete components with simple interactions.

cond-mat.soft↗

arXiv2025

LLM4WM: Adapting LLM for Wireless Multi-Tasking

The wireless channel is fundamental to communication, encompassing numerous tasks collectively referred to as channel-associated tasks. These tasks can leverage joint learning based on channel characteristics to share representations and enhance system design. To capitalize on this advantage, LLM4WM is proposed--a large language model (LLM) multi-task fine-tuning framework specifically tailored for channel-associated tasks. This framework utilizes a Mixture of Experts with Low-Rank Adaptation (MoE-LoRA) approach for multi-task fine-tuning, enabling the transfer of the pre-trained LLM's general knowledge to these tasks. Given the unique characteristics of wireless channel data, preprocessing modules, adapter modules, and multi-task output layers are designed to align the channel data with the LLM's semantic feature space. Experiments on a channel-associated multi-task dataset demonstrate that LLM4WM outperforms existing methodologies in both full-sample and few-shot evaluations, owing to its robust multi-task joint modeling and transfer learning capabilities.

eess.SP↗