arXiv subjects

Hong Sun

Publications and source records attributed to Hong Sun.

At least 19 recordsLinked to original sources

arXiv2026

CoRe: A Continuously Reward-Finetuned LLM Query Rewriter for Multi-Stage Context-Aware Relevance in Web-Scale Video Search

LLM-based query rewriters in production face a tension: the training reward must reflect how the rewrite is consumed by the production ranker, yet the training procedure must be cheap enough to support continuous redeployment as data drifts. We present CoRe (Context Relevance), such a system, redeployed weekly for over five months in a major short-video search engine. Our reward uses the deployed multimodal relevance model as its source and a multiplicative ratio form mirroring the production fusion algebra, closing the simulation-production gap that offline reward proxies leave open. A semi-online Mixed Preference Optimization loop makes this reward affordable at multi-million-instance weekly scale: a DPO-style pairwise objective restricts the gradient pass to a small top-k/bottom-k subset of sampled trajectories, and a phase structure reduces trainer/inference-server parameter syncs from per-step to per-phase. An automated promotion gate over reward-like and stability metrics detected and recovered from a real reward-hacking incident in production. Rewriter output is consumed as parallel relevance signals at recall, rawrank, and finerank without displacing the original signals, bounding rewriter-failure blast radius. Online A/B from two sequential production launches, first deploying the rewriter at finerank, then extending consumption to recall and rawrank, delivers statistically significant reductions in change-query rate on rewrite-impacted queries, with all headline relevance and engagement metrics moving in the expected direction.

cs.IR↗

arXiv2026

The Invitation Trap: Proactive Availability Backdoor in LLMs via Conversational Induction

Current backdoor attacks against LLMs are typically manipulated by the attacker and remain passive. In this paper, we introduce the \textbf{Proactive Availability Backdoor (PAB)}, a novel paradigm that shifts the attack vector from passive waiting to active social engineering. By weaponizing the inherent helpfulness of aligned LLMs, PAB proactively traps users into executing trigger-implanted queries by offering suggestions, achieving high aggressiveness, precision and stealthiness. To rigorously evaluate its threat in a real-life context, we introduce a dual-agent ecological simulation framework based on selected dimensions of the Five-Factor Model, and deploy PAB with few-shot prompts. Being validated on different models and domains, PAB performs remarkably and its effective attack success rate, which calculates the joint probability of attack incidence rate and attack success rate, goes to \textbf{73.1\%}. We also introduce \textbf{Anti-PAB}, a defense method tailored for PAB. Our findings reveal that the helpfulness of LLMs can be weaponized to compromise availability, exposing a serious hidden threat to LLMs users. We release all the scripts and datasets in the experiments at \texttt{https://anonymous.4open.science/r/PAB-ANONYMOUS/}.

cs.CR↗

arXiv2026

Origin of Giant Phonon Magnetic Moment in Orbital Seebeck Effect: a Heisenberg-type L-L Coupling

Inspired by the recent observation of the orbital Seebeck effect in alpha-quartz, we identify an intrinsic amplification mechanism for thermally generated phonon angular momentum and phonon magnetic moment in chiral insulators. We propose a Heisenberg-type long-range coupling between phonon angular momenta, referred to here as L-L coupling, which opens a self-consistent feedback channel and strongly enhances the bare thermal response within linear response. Our calculations reveal a pronounced temperature- and size-dependent amplification, dominated by the off-diagonal channel, with the total phonon angular momentum enhanced by up to nearly two orders of magnitude as the system approaches the threshold from below. These findings suggest that L-L coupling may provide a microscopic origin of giant phonon magnetic moment the recently observed orbital Seebeck effect in alpha-quartz.

cond-mat.other↗

arXiv2025

MCVI-SANet: A lightweight semi-supervised model for LAI and SPAD estimation of winter wheat under vegetation index saturation

Vegetation index (VI) saturation during the dense canopy stage and limited ground-truth annotations of winter wheat constrain accurate estimation of LAI and SPAD. Existing VI-based and texture-driven machine learning methods exhibit limited feature expressiveness. In addition, deep learning baselines suffer from domain gaps and high data demands, which restrict their generalization. Therefore, this study proposes the Multi-Channel Vegetation Indices Saturation Aware Net (MCVI-SANet), a lightweight semi-supervised vision model. The model incorporates a newly designed Vegetation Index Saturation-Aware Block (VI-SABlock) for adaptive channel-spatial feature enhancement. It also integrates a VICReg-based semi-supervised strategy to further improve generalization. Datasets were partitioned using a vegetation height-informed strategy to maintain representativeness across growth stages. Experiments over 10 repeated runs demonstrate that MCVI-SANet achieves state-of-the-art accuracy. The model attains an average R2 of 0.8123 and RMSE of 0.4796 for LAI, and an average R2 of 0.6846 and RMSE of 2.4222 for SPAD. This performance surpasses the best-performing baselines, with improvements of 8.95% in average LAI R2 and 8.17% in average SPAD R2. Moreover, MCVI-SANet maintains high inference speed with only 0.10M parameters. Overall, the integration of semi-supervised learning with agronomic priors provides a promising approach for enhancing remote sensing-based precision agriculture.

cs.CV↗

arXiv2025

Preference is More Than Comparisons: Rethinking Dueling Bandits with Augmented Human Feedback

Interactive preference elicitation (IPE) aims to substantially reduce human effort while acquiring human preferences in wide personalization systems. Dueling bandit (DB) algorithms enable optimal decision-making in IPE building on pairwise comparisons. However, they remain inefficient when human feedback is sparse. Existing methods address sparsity by heavily relying on parametric reward models, whose rigid assumptions are vulnerable to misspecification. In contrast, we explore an alternative perspective based on feedback augmentation, and introduce critical improvements to the model-free DB framework. Specifically, we introduce augmented confidence bounds to integrate augmented human feedback under generalized concentration properties, and analyze the multi-factored performance trade-off via regret analysis. Our prototype algorithm achieves competitive performance across several IPE benchmarks, including recommendation, multi-objective optimization, and response optimization for large language models, demonstrating the potential of our approach for provably efficient IPE in broader applications.

cs.LG↗

arXiv2025

Unsupervised Atomic Data Mining via Multi-Kernel Graph Autoencoders for Machine Learning Force Fields

Constructing a chemically diverse dataset while avoiding sampling bias is critical to training efficient and generalizable force fields. However, in computational chemistry and materials science, many common dataset generation techniques are prone to oversampling regions of the potential energy surface. Furthermore, these regions can be difficult to identify and isolate from each other or may not align well with human intuition, making it challenging to systematically remove bias in the dataset. While traditional clustering and pruning (down-sampling) approaches can be useful for this, they can often lead to information loss or a failure to properly identify distinct regions of the potential energy surface due to difficulties associated with the high dimensionality of atomic descriptors. In this work, we introduce the Multi-kernel Edge Attention-based Graph Autoencoder (MEAGraph) model, an unsupervised approach for analyzing atomic datasets. MEAGraph combines multiple linear kernel transformations with attention-based message passing to capture geometric sensitivity and enable effective dataset pruning without relying on labels or extensive training. Demonstrated applications on niobium, tantalum, and iron datasets show that MEAGraph efficiently groups similar atomic environments, allowing for the use of basic pruning techniques for removing sampling bias. This approach provides an effective method for representation learning and clustering that can be used for data analysis, outlier detection, and dataset optimization.

cs.LG↗

arXiv2025

USCTNet: A deep unfolding nuclear-norm optimization solver for physically consistent HSI reconstruction

Reconstructing hyperspectral images (HSIs) from a single RGB image is ill-posed and can become physically inconsistent when the camera spectral sensitivity (CSS) and scene illumination are misspecified. We formulate RGB-to-HSI reconstruction as a physics-grounded inverse problem regularized by a nuclear norm in a learnable transform domain, and we explicitly estimate CSS and illumination to define the forward operator embedded in each iteration, ensuring colorimetric consistency. To avoid the cost and instability of full singular-value decompositions (SVDs) required by singular-value thresholding (SVT), we introduce a data-adaptive low-rank subspace SVT operator. Building on these components, we develop USCTNet, a deep unfolding solver tailored to HSI that couples a parameter estimation module with learnable proximal updates. Extensive experiments on standard benchmarks show consistent improvements over state-of-the-art RGB-based methods in reconstruction accuracy. Code: https://github.com/psykheXX/USCTNet-Code-Implementation.git

cs.CV↗

arXiv2025

Beyond Label Semantics: Language-Guided Action Anatomy for Few-shot Action Recognition

Few-shot action recognition (FSAR) aims to classify human actions in videos with only a small number of labeled samples per category. The scarcity of training data has driven recent efforts to incorporate additional modalities, particularly text. However, the subtle variations in human posture, motion dynamics, and the object interactions that occur during different phases, are critical inherent knowledge of actions that cannot be fully exploited by action labels alone. In this work, we propose Language-Guided Action Anatomy (LGA), a novel framework that goes beyond label semantics by leveraging Large Language Models (LLMs) to dissect the essential representational characteristics hidden beneath action labels. Guided by the prior knowledge encoded in LLM, LGA effectively captures rich spatiotemporal cues in few-shot scenarios. Specifically, for text, we prompt an off-the-shelf LLM to anatomize labels into sequences of atomic action descriptions, focusing on the three core elements of action (subject, motion, object). For videos, a Visual Anatomy Module segments actions into atomic video phases to capture the sequential structure of actions. A fine-grained fusion strategy then integrates textual and visual features at the atomic level, resulting in more generalizable prototypes. Finally, we introduce a Multimodal Matching mechanism, comprising both video-video and video-text matching, to ensure robust few-shot classification. Experimental results demonstrate that LGA achieves state-of-the-art performance across multipe FSAR benchmarks.

cs.CV↗

arXiv2025

On-the-fly machine learning-augmented constrained AIMD to design new routes from glassy carbon to quenchable amorphous diamond with low pressure and temperature

Recent advances in machine learning have enabled large-scale atomic simulations with first-principles accuracy, allowing precise modeling of disordered materials such as glassy carbon (GC). However, conventional ab initio molecular dynamics (AIMD) cannot effectively capture anisotropic stress effects, which are believed to play a key role in the transformation of GC into amorphous diamond under extreme conditions. In this work, we present an on-the-fly machine learning-augmented constrained AIMD (ML-augmented CAIMD) approach by modifying VASP 6.3.2. Our simulations not only reproduce major experimental features of GC but also provide restrictive synthesis conditions and microscopic insights. We show that GC exhibits unexpectedly high plasticity, with its compressive and shear strengths enhanced by large strains. Under pressure, increasing annealing temperature promotes the formation of quenchable amorphous diamond via enhanced sp3 preservation, but this trend reverses above 2900 K due to thermal graphitization. Under non-hydrostatic compression, GC transforms into a superhard structure sustaining large stress differences, which sharply increase when confining pressure exceeds 40 GPa. Finally, severe rotational shear at 30 GPa induces sp3 fractions up to 80 percent at 300 to 1000 K. A hardened amorphous carbon retaining 64 percent sp3 content is achieved by decompression at 300 K, marking the lowest pressure-temperature route ever predicted. Our ML-augmented CAIMD provides a general framework for modeling structural transformations in disordered materials under anisotropic stresses.

cond-mat.mtrl-sci↗

arXiv2025

Finite-temperature Green's function theory of terahertz-induced phonon angular momentum in polar crystals

We develop a self-energy-dressed Green's-function framework for terahertz-induced phonon angular momentum in polar crystals. The rectified angular-momentum response is formulated as a second-order response and, within a dressed-bubble approximation, is expressed in terms of retarded phonon propagators weighted by mode-resolved polarization and angular-momentum matrix elements. Anharmonic self-energies enter through the dressed propagators, incorporating finite-temperature frequency renormalization and linewidth broadening directly into the response kernel. Applying this framework to wurtzite GaN, we show that the terahertz propagation direction and polarization select distinct rotational phonon channels: a nondegenerate $\mathrm{E}_{1}(\mathrm{TO})$--$\mathrm{A}_{1}(\mathrm{TO})$ channel with phase- and frequency-tunable angular momentum, and a degenerate $\mathrm{E}_{1}(\mathrm{TO})$ channel with helicity-selected response. Anharmonic broadening suppresses and smooths the resonant structures while preserving the characteristic phase dependence of each channel. An order-of-magnitude estimate based on the phonon inverse Faraday-effect framework gives an mT-scale effective magnetic field for representative electron--phonon coupling strengths. This work places coherent terahertz-driven circular-phonon physics in a first-principles-based response framework and provides a starting point for going beyond constant-damping driven-mode descriptions.

cond-mat.mtrl-sci↗

arXiv2025

Ab Initio Theory of Phonon Magnetic Moment Induced by Electron-Phonon Coupling in Magnetic Materials

Circularly polarized phonons, characterized by nonzero angular momenta and magnetic moments, have attracted extensive attention. However, a long-standing critical issue in this field is the lack of an approach to accurately calculate phonon magnetic moments resulting from electron-phonon coupling (EPC) in realistic materials. Here, based on the linear response framework, we develop an ab initio theory for calculating EPC-induced magnetic properties of phonons, applicable to both insulating and metallic materials. Our method can precisely calculate phonon Zeeman splittings in magnetic metals with significant EPC, as demonstrated by the remarkable agreement with recent experimental observations of phonon Zeeman splitting in the ferromagnetic Weyl semimetal Co3Sn2S2. In addition, the long-sought magnetic phonon spectra across the entire Brillouin zone are obtained, facilitating the study of magnetic phonon transport and topology. Specifically, by constructing an inertially decoupled lattice model, we propose candidate materials exhibiting intrinsic phonon Chern states with robust unidirectional edge phonon currents. Our work paves the way for investigating novel phonon phenomena in magnetic quantum materials.

cond-mat.mtrl-sci↗

arXiv2024

Bayesian Nowcasting Data Breach IBNR Incidents

The reporting delay in data breach incidents poses a formidable challenge for Incurred But Not Reported (IBNR) studies, complicating reserve estimation for actuarial professionals. This work presents a novel Bayesian nowcasting model designed to accurately model and predict the number of IBNR data breach incidents. Leveraging a Bayesian modeling framework, the model integrates time and heterogeneous effects to enhance predictive accuracy. Synthetic and empirical studies demonstrate the superior performance of the proposed model, highlighting its efficacy in addressing the complexities of IBNR estimation. Furthermore, we examine reserve estimation for IBNR incidents using the proposed model, shedding light on its implications for actuarial practice.

stat.AP↗

arXiv2024

Ice phase classification made easy with score-based denoising

Accurate identification of ice phases is essential for understanding various physicochemical phenomena. However, such classification for structures simulated with molecular dynamics is complicated by the complex symmetries of ice polymorphs and thermal fluctuations. For this purpose, both traditional order parameters and data-driven machine learning approaches have been employed, but they often rely on expert intuition, specific geometric information, or large training datasets. In this work, we present an unsupervised phase classification framework that combines a score-based denoiser model with a subsequent model-free classification method to accurately identify ice phases. The denoiser model is trained on perturbed synthetic data of ideal reference structures, eliminating the need for large datasets and labeling efforts. The classification step utilizes the Smooth Overlap of Atomic Positions (SOAP) descriptors as the atomic fingerprint, ensuring Euclidean symmetries and transferability to various structural systems. Our approach achieves a remarkable 100\% accuracy in distinguishing ice phases of test trajectories using only seven ideal reference structures of ice phases as model inputs. This demonstrates the generalizability of the score-based denoiser model in facilitating phase identification for complex molecular systems. The proposed classification strategy can be broadly applied to investigate structural evolution and phase identification for a wide range of materials, offering new insights into the fundamental understanding of water and other complex systems.

cond-mat.mtrl-sci↗

arXiv2024

Real is not True: Backdoor Attacks Against Deepfake Detection

The proliferation of malicious deepfake applications has ignited substantial public apprehension, casting a shadow of doubt upon the integrity of digital media. Despite the development of proficient deepfake detection mechanisms, they persistently demonstrate pronounced vulnerability to an array of attacks. It is noteworthy that the pre-existing repertoire of attacks predominantly comprises adversarial example attack, predominantly manifesting during the testing phase. In the present study, we introduce a pioneering paradigm denominated as Bad-Deepfake, which represents a novel foray into the realm of backdoor attacks levied against deepfake detectors. Our approach hinges upon the strategic manipulation of a delimited subset of the training data, enabling us to wield disproportionate influence over the operational characteristics of a trained model. This manipulation leverages inherent frailties inherent to deepfake detectors, affording us the capacity to engineer triggers and judiciously select the most efficacious samples for the construction of the poisoned set. Through the synergistic amalgamation of these sophisticated techniques, we achieve an remarkable performance-a 100% attack success rate (ASR) against extensively employed deepfake detectors.

cs.CR↗

arXiv2023

Asymptotically compatible energy and dissipation law of the nonuniform L2-$1_{\sigma}$ scheme for time fractional Allen-Cahn model

We build an asymptotically compatible energy of the variable-step L2-$1_{\sigma}$ scheme for the time-fractional Allen-Cahn model with the Caputo's fractional derivative of order $\alpha\in(0,1)$, under a weak step-ratio constraint $\tau_k/\tau_{k-1}\geq r_{\star}(\alpha)$ for $k\ge2$, where $\tau_k$ is the $k$-th time-step size and $r_{\star}(\alpha)\in(0.3865,0.4037)$ for $\alpha\in(0,1)$. It provides a positive answer to the open problem in [J. Comput. Phys., 414:109473], and, to the best of our knowledge, it is the first second-order nonuniform time-stepping scheme to preserve both the maximum bound principle and the energy dissipation law of time-fractional Allen-Cahn model. The compatible discrete energy is constructed via a novel discrete gradient structure of the second-order L2-$1_{\sigma}$ formula by a local-nonlocal splitting technique. It splits the discrete fractional derivative into two parts: one is a local term analogue to the trapezoid rule of the first derivative and the other is a nonlocal summation analogue to the L1 formula of Caputo derivative. Numerical examples with an adaptive time-stepping strategy are provided to show the effectiveness of our scheme and the asymptotic properties of the associated modified energy.

math.NA↗

arXiv2023

Novel models for fatigue life prediction under wideband random loads based on machine learning

Machine learning as a data-driven solution has been widely applied in the field of fatigue lifetime prediction. In this paper, three models for wideband fatigue life prediction are built based on three machine learning models, i.e. support vector machine (SVM), Gaussian process regression (GPR) and artificial neural network (ANN). The generalization ability of the models is enhanced by employing numerous power spectra samples with different bandwidth parameters and a variety of material properties related to fatigue life. Sufficient Monte Carlo numerical simulations demonstrate that the newly developed machine learning models are superior to the traditional frequency-domain models in terms of life prediction accuracy and the ANN model has the best overall performance among the three developed machine learning models.

cond-mat.mtrl-sci↗

arXiv2023

Explore the Effect of Data Selection on Poison Efficiency in Backdoor Attacks

As the number of parameters in Deep Neural Networks (DNNs) scales, the thirst for training data also increases. To save costs, it has become common for users and enterprises to delegate time-consuming data collection to third parties. Unfortunately, recent research has shown that this practice raises the risk of DNNs being exposed to backdoor attacks. Specifically, an attacker can maliciously control the behavior of a trained model by poisoning a small portion of the training data. In this study, we focus on improving the poisoning efficiency of backdoor attacks from the sample selection perspective. The existing attack methods construct such poisoned samples by randomly selecting some clean data from the benign set and then embedding a trigger into them. However, this random selection strategy ignores that each sample may contribute differently to the backdoor injection, thereby reducing the poisoning efficiency. To address the above problem, a new selection strategy named Improved Filtering and Updating Strategy (FUS++) is proposed. Specifically, we adopt the forgetting events of the samples to indicate the contribution of different poisoned samples and use the curvature of the loss surface to analyses the effectiveness of this phenomenon. Accordingly, we combine forgetting events and curvature of different samples to conduct a simple yet efficient sample selection strategy. The experimental results on image classification (CIFAR-10, CIFAR-100, ImageNet-10), text classification (AG News), audio classification (ESC-50), and age regression (Facial Age) consistently demonstrate the effectiveness of the proposed strategy: the attack performance using FUS++ is significantly higher than that using random selection for the same poisoning ratio.

cs.CR↗

arXiv2023

AutoHint: Automatic Prompt Optimization with Hint Generation

This paper presents AutoHint, a novel framework for automatic prompt engineering and optimization for Large Language Models (LLM). While LLMs have demonstrated remarkable ability in achieving high-quality annotation in various tasks, the key to applying this ability to specific tasks lies in developing high-quality prompts. Thus we propose a framework to inherit the merits of both in-context learning and zero-shot learning by incorporating enriched instructions derived from input-output demonstrations to optimize original prompt. We refer to the enrichment as the hint and propose a framework to automatically generate the hint from labeled data. More concretely, starting from an initial prompt, our method first instructs a LLM to deduce new hints for selected samples from incorrect predictions, and then summarizes from per-sample hints and adds the results back to the initial prompt to form a new, enriched instruction. The proposed method is evaluated on the BIG-Bench Instruction Induction dataset for both zero-shot and few-short prompts, where experiments demonstrate our method is able to significantly boost accuracy for multiple tasks.

cs.CL↗