arXiv subjects

Xinyan Wang

Publications and source records attributed to Xinyan Wang.

At least 19 recordsLinked to original sources

arXiv2026

When Are Teacher Tokens Reliable? Position-Weighted On-Policy Self-Distillation for Reasoning

On-policy self-distillation (OPSD) trains a student on its own rollouts using a privileged teacher, but its standard objective weights all generated tokens equally, implicitly treating the privileged teacher target as equally reliable at every student-visited prefix. Existing entropy-based OPD methods relax this uniformity by modulating token-level supervision with teacher entropy, but high teacher entropy in reasoning has an ambiguous reliability meaning: it can reflect either non-viable uncertainty or benign solution diversity. To identify this phenomenon, we introduce a branch-viability diagnostic. Specifically, we record next-token alternatives from the privileged-answer teacher prompt, force each alternative after the student prompt plus its on-policy spine prefix, and test whether the resulting student-template continuation recovers the correct answer. On Qwen3-4B, we find that an oriented within-sequence position score is the strongest tested predictor of teacher-token reliability, reaching an area-under-ROC-curve (AUROC) of 0.83; local uncertainty scores are at most 0.57. Motivated by this trajectory-level structure, we propose Position-Weighted On-Policy Self-Distillation (PW-OPSD), which applies an increasing position weight while keeping the same student rollout, privileged teacher pass, and clipped forward-KL target as OPSD. In our comprehensive evaluations with different random seeds, the diagnostic-derived PW-OPSD improves AIME 2024 and AIME 2025 Avg@12 by +1.0 and +1.1 points, and a generalization evaluation on two larger-scale models from different families, DeepSeek-R1-Distill-Llama-8B and Olmo-3-7B-Think, also demonstrates consistent aggregate Avg@12 improvements. These results show that teacher-token reliability in reasoning distillation is trajectory-structured and can be utilized without additional teacher computation.

cs.LG↗

arXiv2026

ROM: Real-time Overthinking Mitigation via Streaming Detection and Intervention

Large Reasoning Models (LRMs) often reach a correct solution before their long Chain-of-Thought trace ends, yet continue with redundant verification, repeated attempts, or unnecessary exploration that wastes computation and can even overturn the correct answer. We frame this behavior as a latent productive-to-redundant transition and show it is directly reflected in hidden states: around first-correct-solution (FCS) boundaries, late-layer representations separate efficient from overthinking tokens, while boundary-permutation and position controls collapse. We propose ROM, a streaming intervention framework that monitors a frozen LRM with a lightweight hidden-state detector ($\sim$0.1\% of backbone parameters) and intervenes at well-formed reasoning boundaries; Counterfactual Self-Correction (CSC) balances supervision with wrong$\rightarrow$correct trajectories, preserving useful pre-FCS self-correction. Unlike prior adaptive early-exit methods, ROM extracts no intermediate answers, launches no probe decoding, and updates no backbone weights. Across five backbones from three model families and five reasoning benchmarks, against ten recent baselines under a shared protocol, ROM$_{\text{CSC}}$ attains the highest accuracy in 19 of 25 model--benchmark settings, cuts response length by 28--77\% (mean 45\%) versus vanilla decoding, and is the only method on the accuracy--length Pareto front in every setting. The same MATH500-trained supervision transfers zero-shot across scales, families, and task domains, and end-to-end wall-clock latency drops by 46.5\% with $\sim$5\% per-token overhead. Code is available at https://github.com/SaFo-Lab/ROM.

cs.LG↗

arXiv2026

ReasoningBomb: A Stealthy Denial-of-Service Attack by Inducing Pathologically Long Reasoning in Large Reasoning Models

Large reasoning models (LRMs) extend large language models with explicit multi-step reasoning traces, but this capability introduces a new class of prompt-induced inference-time denial-of-service (PI-DoS) attacks that exploit the high computational cost of reasoning. We first formalize inference cost for LRMs and define PI-DoS, then prove that any practical PI-DoS attack should satisfy three properties: (1) a high amplification ratio, where each query induces a disproportionately long reasoning trace relative to its own length; (ii) stealthiness, in which prompts and responses remain on the natural language manifold and evade distribution shift detectors; and (iii) optimizability, in which the attack supports efficient optimization without being slowed by its own success. Under this framework, we present ReasoningBomb, a reinforcement-learning-based PI-DoS framework that is guided by a constant-time surrogate reward and trains a large reasoning-model attacker to generate short natural prompts that drive victim LRMs into pathologically long and often effectively non-terminating reasoning. Across seven open-source models (including LLMs and LRMs) and three commercial LRMs, ReasoningBomb induces 18,759 completion tokens on average and 19,263 reasoning tokens on average across reasoning models. It outperforms the the runner-up baseline by 35% in completion tokens and 38% in reasoning tokens, while inducing 6-7x more tokens than benign queries and achieving 286.7x input-to-output amplification ratio averaged across all samples. Additionally, our method achieves 99.8% bypass rate on input-based detection, 98.7% on output-based detection, and 98.4% against strict dual-stage joint detection.

cs.CR↗

arXiv2025

Effective Probabilistic Time Series Forecasting with Fourier Adaptive Noise-Separated Diffusion

We propose the Fourier Adaptive Lite Diffusion Architecture (FALDA), a novel probabilistic framework for time series forecasting. First, we introduce the Diffusion Model for Residual Regression (DMRR) framework, which unifies diffusion-based probabilistic regression methods. Within this framework, FALDA leverages Fourier-based decomposition to incorporate a component-specific architecture, enabling tailored modeling of individual temporal components. A conditional diffusion model is utilized to estimate the future noise term, while our proposed lightweight denoiser, DEMA (Decomposition MLP with AdaLN), conditions on the historical noise term to enhance denoising performance. Through mathematical analysis and empirical validation, we demonstrate that FALDA effectively reduces epistemic uncertainty, allowing probabilistic learning to primarily focus on aleatoric uncertainty. Experiments on six real-world benchmarks demonstrate that FALDA consistently outperforms existing probabilistic forecasting approaches across most datasets for long-term time series forecasting while achieving enhanced computational efficiency without compromising accuracy. Notably, FALDA also achieves superior overall performance compared to state-of-the-art (SOTA) point forecasting approaches, with improvements of up to 9%.

cs.LG↗

arXiv2025

STCKGE:Continual Knowledge Graph Embedding Based on Spatial Transformation

Current Continual Knowledge Graph Embedding (CKGE) methods primarily rely on translation-based embedding approaches, leveraging previously acquired knowledge to initialize new facts. While these methods often integrate fine-tuning or continual learning strategies to enhance efficiency, they compromise prediction accuracy and lack support for complex relational structures (e.g., multi-hop relations). To address these limitations, we propose STCKGE, a novel CKGE framework based on spatial transformation. In this framework, entity positions are jointly determined by base position vectors and offset vectors, enabling the model to represent complex relations more effectively while supporting efficient embedding updates for both new and existing knowledge through simple spatial operations, without relying on traditional continual learning techniques. Furthermore, we introduce a bidirectional collaborative update strategy and a balanced embedding method to guide parameter updates, effectively minimizing training costs while improving model accuracy. We comprehensively evaluate our model on seven public datasets and a newly constructed dataset (MULTI) focusing on multi-hop relationships. Experimental results confirm STCKGE's strong performance in multi-hop relationship learning and prediction accuracy, with an average MRR improvement of 5.4\%.

cs.IR↗

arXiv2024

On the design space between molecular mechanics and machine learning force fields

A force field as accurate as quantum mechanics (QM) and as fast as molecular mechanics (MM), with which one can simulate a biomolecular system efficiently enough and meaningfully enough to get quantitative insights, is among the most ardent dreams of biophysicists -- a dream, nevertheless, not to be fulfilled any time soon. Machine learning force fields (MLFFs) represent a meaningful endeavor towards this direction, where differentiable neural functions are parametrized to fit ab initio energies, and furthermore forces through automatic differentiation. We argue that, as of now, the utility of the MLFF models is no longer bottlenecked by accuracy but primarily by their speed (as well as stability and generalizability), as many recent variants, on limited chemical spaces, have long surpassed the chemical accuracy of $1$ kcal/mol -- the empirical threshold beyond which realistic chemical predictions are possible -- though still magnitudes slower than MM. Hoping to kindle explorations and designs of faster, albeit perhaps slightly less accurate MLFFs, in this review, we focus our attention on the design space (the speed-accuracy tradeoff) between MM and ML force fields. After a brief review of the building blocks of force fields of either kind, we discuss the desired properties and challenges now faced by the force field development community, survey the efforts to make MM force fields more accurate and ML force fields faster, envision what the next generation of MLFF might look like.

physics.chem-ph↗

arXiv2024

ISPO: An Integrated Ontology of Symptom Phenotypes for Semantic Integration of Traditional Chinese Medical Data

Symptom phenotypes are one of the key types of manifestations for diagnosis and treatment of various disease conditions. However, the diversity of symptom terminologies is one of the major obstacles hindering the analysis and knowledge sharing of various types of symptom-related medical data particularly in the fields of Traditional Chinese Medicine (TCM). Objective: This study aimed to construct an Integrated Ontology of symptom phenotypes (ISPO) to support the data mining of Chinese EMRs and real-world study in TCM field. Methods: To construct an integrated ontology of symptom phenotypes (ISPO), we manually annotated classical TCM textbooks and large-scale Chinese electronic medical records (EMRs) to collect symptom terms with support from a medical text annotation system. Furthermore, to facilitate the semantic interoperability between different terminologies, we incorporated public available biomedical vocabularies by manual mapping between Chinese terms and English terms with cross-references to source vocabularies. In addition, we evaluated the ISPO using independent clinical EMRs to provide a high-usable medical ontology for clinical data analysis. Results: By integrating 78,696 inpatient cases of EMRs, 5 biomedical vocabularies, 21 TCM books and dictionaries, ISPO provides 3,147 concepts, 23,475 terms, and 55,552 definition or contextual texts. Adhering to the taxonomical structure of the related anatomical systems of symptom phenotypes, ISPO provides 12 top-level categories and 79 middle-level sub-categories. The validation of data analysis showed the ISPO has a coverage rate of 95.35%, 98.53% and 92.66% for symptom terms with occurrence rates of 0.5% in additional three independent curated clinical datasets, which can demonstrate the significant value of ISPO in mapping clinical terms to ontologies.

cs.CL↗

arXiv2024

Observation of dendrite formation at Li metal-electrolyte interface: A machine-learning enhanced constant potential framework

Uncontrollable dendrites growth during electrochemical cycles leads to low Coulombic efficiency and critical safety issues in Li metal batteries. Hence, a comprehensive understanding of the dendrite formation mechanism is essential for further enhancing the performance of Li metal batteries. Machine learning accelerated molecular dynamics (MD) simulations can provide atomic-scale resolution for various key processes at an ab-initio level accuracy. However, traditional MD simulation tools hardly capture Li electrochemical depositions, due to lack of an electrochemical constant potential (ConstP) condition. In this work, we propose a ConstP approach that combines a machine learning force field with the charge equilibration method to reveal the dynamic process of dendrites nucleation at Li metal anode surfaces. Our simulations show that inhomogeneous Li depositions, following Li aggregations in amorphous inorganic components of solid electrolyte interphases, can initiate dendrites nucleation. Our study provides microscopic insights for Li dendrites formations in Li metal anodes. More importantly, we present an efficient and accurate simulation method for modeling realistic ConstP conditions, which holds considerable potential for broader applications in modeling complex electrochemical interfaces.

cond-mat.mtrl-sci↗

arXiv2024

Efficient and Precise Force Field Optimization for Biomolecules Using DPA-2

Molecular simulations are essential tools in computational chemistry, enabling the prediction and understanding of molecular interactions and thermodynamic properties of biomolecules. However, traditional force fields face significant challenges in accurately representing novel molecules and complex chemical environments due to the labor-intensive process of manually setting optimization parameters and the high computational cost of quantum mechanical calculations. To overcome these difficulties, we fine-tuned a high-accuracy DPA-2 pre-trained model and applied it to optimize force field parameters on-the-fly, significantly reducing computational costs. Our method combines this fine-tuned DPA-2 model with a node-embedding-based similarity metric, allowing seamless augmentation to new chemical species without manual intervention. We applied this process to the TYK2 inhibitor and PTP1B systems and demonstrated its effectiveness through the improvement of free energy perturbation calculation results. This advancement contributes valuable insights and tools for the computational chemistry community.

physics.chem-ph↗

arXiv2024

DDPM-MoCo: Advancing Industrial Surface Defect Generation and Detection with Generative and Contrastive Learning

The task of industrial detection based on deep learning often involves solving two problems: (1) obtaining sufficient and effective data samples, (2) and using efficient and convenient model training methods. In this paper, we introduce a novel defect-generation method, named DDPM-MoCo, to address these issues. Firstly, we utilize the Denoising Diffusion Probabilistic Model (DDPM) to generate high-quality defect data samples, overcoming the problem of insufficient sample data for model learning. Furthermore, we utilize the unsupervised learning Momentum Contrast model (MoCo) with an enhanced batch contrastive loss function for training the model on unlabeled data, addressing the efficiency and consistency challenges in large-scale negative sample encoding during diffusion model training. The experimental results showcase an enhanced visual detection method for identifying defects on metal surfaces, covering the entire process, starting from generating unlabeled sample data for training the diffusion model, to utilizing the same labeled sample data for downstream detection tasks. This study offers valuable practical insights and application potential for visual detection in the metal processing industry.

cs.CV↗

arXiv2024

ADMarker: A Multi-Modal Federated Learning System for Monitoring Digital Biomarkers of Alzheimer's Disease

Alzheimer's Disease (AD) and related dementia are a growing global health challenge due to the aging population. In this paper, we present ADMarker, the first end-to-end system that integrates multi-modal sensors and new federated learning algorithms for detecting multidimensional AD digital biomarkers in natural living environments. ADMarker features a novel three-stage multi-modal federated learning architecture that can accurately detect digital biomarkers in a privacy-preserving manner. Our approach collectively addresses several major real-world challenges, such as limited data labels, data heterogeneity, and limited computing resources. We built a compact multi-modality hardware system and deployed it in a four-week clinical trial involving 91 elderly participants. The results indicate that ADMarker can accurately detect a comprehensive set of digital biomarkers with up to 93.8% accuracy and identify early AD with an average of 88.9% accuracy. ADMarker offers a new platform that can allow AD clinicians to characterize and track the complex correlation between multidimensional interpretable digital biomarkers, demographic factors of patients, and AD diagnosis in a longitudinal manner.

cs.LG↗

arXiv2024

Graph Neural Aggregation-diffusion with Metastability

Continuous graph neural models based on differential equations have expanded the architecture of graph neural networks (GNNs). Due to the connection between graph diffusion and message passing, diffusion-based models have been widely studied. However, diffusion naturally drives the system towards an equilibrium state, leading to issues like over-smoothing. To this end, we propose GRADE inspired by graph aggregation-diffusion equations, which includes the delicate balance between nonlinear diffusion and aggregation induced by interaction potentials. The node representations obtained through aggregation-diffusion equations exhibit metastability, indicating that features can aggregate into multiple clusters. In addition, the dynamics within these clusters can persist for long time periods, offering the potential to alleviate over-smoothing effects. This nonlinear diffusion in our model generalizes existing diffusion-based models and establishes a connection with classical GNNs. We prove that GRADE achieves competitive performance across various benchmarks and alleviates the over-smoothing issue in GNNs evidenced by the enhanced Dirichlet energy.

cs.LG↗

arXiv2024

Understanding Oversmoothing in Diffusion-Based GNNs From the Perspective of Operator Semigroup Theory

This paper presents an analytical study of the oversmoothing issue in diffusion-based Graph Neural Networks (GNNs). Generalizing beyond extant approaches grounded in random walk analysis or particle systems, we approach this problem through operator semigroup theory. This theoretical framework allows us to rigorously prove that oversmoothing is intrinsically linked to the ergodicity of the diffusion operator. Relying on semigroup method, we can quantitatively analyze the dynamic of graph diffusion and give a specific mathematical form of the smoothing feature by ergodicity and invariant measure of operator, which improves previous works only show existence of oversmoothing. This finding further poses a general and mild ergodicity-breaking condition, encompassing the various specific solutions previously offered, thereby presenting a more universal and theoretically grounded approach to relieve oversmoothing in diffusion-based GNNs. Additionally, we offer a probabilistic interpretation of our theory, forging a link with prior works and broadening the theoretical horizon. Our experimental results reveal that this ergodicity-breaking term effectively mitigates oversmoothing measured by Dirichlet energy, and simultaneously enhances performance in node classification tasks.

cs.LG↗

arXiv2023

DPA-2: a large atomic model as a multi-task learner

The rapid advancements in artificial intelligence (AI) are catalyzing transformative changes in atomic modeling, simulation, and design. AI-driven potential energy models have demonstrated the capability to conduct large-scale, long-duration simulations with the accuracy of ab initio electronic structure methods. However, the model generation process remains a bottleneck for large-scale applications. We propose a shift towards a model-centric ecosystem, wherein a large atomic model (LAM), pre-trained across multiple disciplines, can be efficiently fine-tuned and distilled for various downstream tasks, thereby establishing a new framework for molecular modeling. In this study, we introduce the DPA-2 architecture as a prototype for LAMs. Pre-trained on a diverse array of chemical and materials systems using a multi-task approach, DPA-2 demonstrates superior generalization capabilities across multiple downstream tasks compared to the traditional single-task pre-training and fine-tuning methodologies. Our approach sets the stage for the development and broad application of LAMs in molecular and materials simulation research.

physics.chem-ph↗

arXiv2023

FFF: Fragments-Guided Flexible Fitting for Building Complete Protein Structures

Cryo-electron microscopy (cryo-EM) is a technique for reconstructing the 3-dimensional (3D) structure of biomolecules (especially large protein complexes and molecular assemblies). As the resolution increases to the near-atomic scale, building protein structures de novo from cryo-EM maps becomes possible. Recently, recognition-based de novo building methods have shown the potential to streamline this process. However, it cannot build a complete structure due to the low signal-to-noise ratio (SNR) problem. At the same time, AlphaFold has led to a great breakthrough in predicting protein structures. This has inspired us to combine fragment recognition and structure prediction methods to build a complete structure. In this paper, we propose a new method named FFF that bridges protein structure prediction and protein structure recognition with flexible fitting. First, a multi-level recognition network is used to capture various structural features from the input 3D cryo-EM map. Next, protein structural fragments are generated using pseudo peptide vectors and a protein sequence alignment method based on these extracted features. Finally, a complete structural model is constructed using the predicted protein fragments via flexible fitting. Based on our benchmark tests, FFF outperforms the baseline methods for building complete protein structures.

cs.CV↗

arXiv2023

A Pre-training Framework for Knowledge Graph Completion

Knowledge graph completion (KGC) is one of the effective methods to identify new facts in knowledge graph. Except for a few methods based on graph network, most of KGC methods trend to be trained based on independent triples, while are difficult to take a full account of the information of global network connection contained in knowledge network. To address these issues, in this study, we propose a simple and effective Network-based Pre-training framework for knowledge graph completion (termed NetPeace), which takes into account the information of global network connection and local triple relationships in knowledge graph. Experiments show that in NetPeace framework, multiple KGC models yields consistent and significant improvements on benchmarks (e.g., 36.45% Hits@1 and 27.40% MRR improvements for TuckER on FB15k-237), especially dense knowledge graph. On the challenging low-resource task, NetPeace that benefits from the global features of KG achieves higher performance (104.03% MRR and 143.89% Hit@1 improvements at most) than original models.

cs.AI↗

arXiv2023

Knowledge Graph Completion based on Tensor Decomposition for Disease Gene Prediction

Accurate identification of disease genes has consistently been one of the keys to decoding a disease's molecular mechanism. Most current approaches focus on constructing biological networks and utilizing machine learning, especially, deep learning to identify disease genes, but ignore the complex relations between entities in the biological knowledge graph. In this paper, we construct a biological knowledge graph centered on diseases and genes, and develop an end-to-end Knowledge graph completion model for Disease Gene Prediction using interactional tensor decomposition (called KDGene). KDGene introduces an interaction module between the embeddings of entities and relations to tensor decomposition, which can effectively enhance the information interaction in biological knowledge. Experimental results show that KDGene significantly outperforms state-of-the-art algorithms. Furthermore, the comprehensive biological analysis of the case of diabetes mellitus confirms KDGene's ability for identifying new and accurate candidate genes. This work proposes a scalable knowledge graph completion framework to identify disease candidate genes, from which the results are promising to provide valuable references for further wet experiments.

cs.AI↗

arXiv2022

Single charge control of localized excitons in heterostructures with ferroelectric thin films and two-dimensional transition metal dichalcogenides

Single charge control of localized excitons (LXs) in two-dimensional transition metal dichalcogenides (TMDCs) is crucial for potential applications in quantum information processing and storage. However, traditional electrostatic doping method with applying metallic gates onto TMDCs may cause the inhomogeneous charge distribution, optical quench, and energy loss. Here, by locally controlling the ferroelectric polarization of the ferroelectric thin film BiFeO3 (BFO) with a scanning probe, we can deterministically manipulate the doping type of monolayer WSe2 to achieve the p-type and n-type doping. This nonvolatile approach can maintain the doping type and hold the localized excitonic charges for a long time without applied voltage. Our work demonstrated that ferroelectric polarization of BFO can control the charges of LXs effectively. Neutral and charged LXs have been observed in different ferroelectric polarization regions, confirmed by magnetic optical measurement. Highly circular polarization degree about 90 % of the photon emission from these quantum emitters have been achieved in high magnetic fields. Controlling single charge of LXs in a non-volatile way shows a great potential for deterministic photon emission with desired charge states for photonic long-term memory.

cond-mat.mes-hall↗