Search arXiv⌕ Search

arXiv subjects

Liang Dong

Publications and source records attributed to Liang Dong.

At least 19 recordsLinked to original sources
arXiv2026

What Gets Hacked in a Collaborative Reward for LLM Recommenders: The Part That Does Not Depend on the User

Collaborative-filtering (CF) scores provide graded, personalized rewards for training large language model (LLM) recommenders through reinforcement learning. However, these rewards can cause exposure collapse: even though the CF model itself produces diverse top recommendations across users, an LLM optimized with its rewards concentrates on fewer items. We decompose each score into an item-mean component, the item's average CF score across user histories, and a residual component, the current user's score deviation from the average. Through controlled interventions, we identify the item-mean component as a contributor to exposure collapse. Accordingly, we propose two reward corrections: per-item centering removes each item's estimated mean score, while per-item standardization further rescales the residual using its estimated standard deviation. Experiments on two datasets with two identifier schemes show that both corrections substantially improve catalog coverage and exposure evenness. Compared with training on the uncorrected CF reward, our corrections increase the number of distinct items ranked first across users by up to $9.8\times$, while maintaining comparable ranking accuracy. Code is available at https://anonymous.4open.science/r/LLM4Rec-CF-Bias-27EF/.

cs.IR↗
arXiv2026

When Unlearning Fails: Reliable Data Deletion under Post-Training in Agent Networks

Embedded AI agents on edge devices can learn as a federation and keep improving after deployment. Each agent collects new trajectories with the current policy, and the edge server aggregates them in later rounds. In such self-improving agent networks, deleting data is harder than a one-time model repair. The data to be deleted may have already shaped later retained trajectories, so retraining or model-side unlearning leaves an influence echo that returns as the network keeps learning. We propose Muting Unlearned Trajectories' Echoes (MUTE) for reliable deletion over edge networks. MUTE traces the echo from a lightweight ledger at the edge server, without moving raw trajectories off the devices. It erases the current residue with a forget-retain update, contains high-influence retained trajectories, and audits later behavior to schedule further erasure under the uplink budget and the response deadline. On LIBERO, MUTE keeps the regenerated share of forgotten behavior at or below 0.401 for every request type, against at least 0.466 for five recent methods.

cs.NI↗
arXiv2026

MemCalib: Benchmarking and Optimizing Memory Use in LLM Agents

The effectiveness of agent memory ultimately depends on whether the underlying LLM gives each memory in context an appropriate degree of influence over its response. Yet this capability has remained largely overlooked. To assess this capability, we introduce MemCalib, a benchmark grounded in realistic memory-system scenarios for evaluating memory use and advancing optimization algorithms. Results on the MemCalib test set reveal that frontier open- and closed-source models struggle to use memory appropriately. They frequently over-use or under-use memory rather than matching each proposition's actual use to its target level, leading to biased, low-quality responses. Experiments with common post-training algorithms, including group relative policy optimization and on-policy self-distillation, further reveal a clear directional skew: trained models improve in one direction while deteriorating in the other. We therefore propose MemCalib-RL, an ordered bidirectional counterfactual credit-assignment algorithm that separates over- and under-use signals and localizes their credit to response tokens through exact atom ablation. Results across model families and scales (Qwen3-8B, Ministral-3-8B-Instruct, and Qwen3.5-35B-A3B) show that MemCalib-RL achieves the best overall performance while better balancing over-use and under-use, with gains generalizing beyond MemCalib in external benchmark evaluation. Further experiments support its design choices and robustness and provide insight into its training dynamics.

cs.LG↗
arXiv2026

Data-driven measurements of the beauty hadron production in p-Pb collisions at \sqrt{s_{\mathrm{NN}}} = 5.02 and 8.16 TeV

Beauty-hadron production in proton-nucleus collisions is a particularly powerful probe of cold nuclear matter (CNM) effects: thanks to the large beauty-quark mass, the production is initiated in early hard scatterings and is calculable in perturbative QCD, while the modifications observed in nuclear collisions directly constrain the nuclear parton distribution functions (nPDFs) in a kinematic region complementary to that accessed by charm. Precisely quantifying these effects is a prerequisite for isolating possible signatures of quark-gluon plasma formation in heavy-ion collisions. In this work, the production of open beauty hadrons in p-Pb collisions at $\sqrt{s_{\mathrm{NN}}} = 5.02$ and $8.16$~TeV is studied via a data-driven Bayesian unfolding of ALICE and LHCb measurements of non-prompt $D^{0}$ and J/$ψ$ mesons. The unfolded $p_{T}$- and $y$-differential cross sections are mutually consistent across decay channels and compatible with directly reconstructed beauty-hadron measurements. The nuclear modification factor $R_{pPb}$ is reported as a function of transverse momentum and rapidity, providing the first unfolding-based result spanning $-4.0 < y < 4.0$ of beauty-hadron $R_{pPb}$ covering the full rapidity range, with a precision comparable to or better than that of existing data. The results are compared with calculations based on nPDFs and can provide valuable constraints for future global nPDF fits. The total inclusive B hadron production cross section in p-Pb collisions is also reported. A weaker CNM suppression is observed for beauty hadrons than for charm hadrons, consistent with the expected mass hierarchy of cold-nuclear-matter effects.

nucl-ex↗
arXiv2026

RPMem: Learning Long-Term Recurrent Parametric Memory Across Sessions for LLM Agents

Long-running LLM agents require memory that persists and evolves across sessions. Text-based memory retrieves and reconstructs past interactions at every query, making long-horizon performance increasingly dependent on retrieval quality and contextual reasoning as histories grow. Parametric memory encodes experience directly into model computation, but existing approaches provide limited support for cross-session memory evolution. Their coupling to a specific backbone further restricts memory reuse after model replacement. We introduce RPMem, a two-stage architecture that compiles each session into a model-independent latent memory through forward computation and selectively integrates it with retained memory via a task-trained recurrent gate. The consolidated memory is then mapped to backbone-specific low-rank adaptation (LoRA) parameters, allowing the encoding capability to transfer when the backbone is replaced. Evaluation across three long-term memory benchmarks and five diverse backbones demonstrates broad generalization with near-constant update cost and memory footprint. With Qwen3-8B on PERMA, RPMem reaches 85.52%, outperforming the strongest parametric and text-based baselines by 5.32 and 12.98 percentage points, respectively. Ablations validate the complementary roles of session compilation and cross-session consolidation, while dynamics analyses reveal that the gate acquires task-specific memory integration strategies. These results establish RPMem as a lifecycle-independent parametric memory framework that maintains evolving cross-session memory that remains reusable across backbone replacements. Our implementation is available at https://github.com/Quark-Medical/rpmem/tree/main.

cs.CL↗
arXiv2026

COPE: Continual Personalization of LLMs under Sparse User Feedback via User Embeddings and Self-Evaluation

While Large Language Models (LLMs) have achieved remarkable results across various benchmarks, their alignment with normative values often results in homogenized responses that fail to address diverse user preferences. Existing training-free methods often occupy valuable context windows through prompt engineering, while training-based methods typically remain static post-training, failing to support the continual optimization required in real-world settings. To address these challenges, we propose COPE (Continual Optimization with Personalized embedding and self-Evaluation), a novel optimization framework tailored for real-world-motivated interaction settings with sparse user feedback. Our framework assigns learnable personalized embeddings to each user and synergistically integrates preference capture, self-evaluation calibration, and personalized response optimization within a single update step. A key innovation of our method is the use of self-evaluation to generate proxy rewards, enabling continuous model updates even when explicit user feedback is unavailable. Experiments show that COPE consistently outperforms strong training-free and training-based baselines under sparse feedback, and remains complementary to Retrieval-Augmented Prompting (RAP). Further analyses confirm COPE's reliable self-evaluation, meaningful preference patterns, stable general capabilities, and robustness under shifting preferences and alternative evaluators.

cs.LG↗
arXiv2026

Atom-light hybrid interferometer for atomic sensing with quantum memory

Quantum memories feature a reversible conversion of optical fields into long-lived atomic spin waves, and are therefore ideal for operating as sensitive atomic sensors. However, up to now, atom-light interferometers have lacked an efficient approach to exploit their ultimate atomic sensing performance, since an extra optical delay line is required to compensate for the memory time. Here, we report a new protocol that records the photocurrent via heterodyne mixing with a stable local oscillator. The obtained complex quadrature amplitude that carries information imprinted on its phase by an external magnetic field, is successfully recovered from the interference patterns between the light and the atomic spin wave, without the stringent requirement of having them overlap in time. Our results reveal that the sensitivity scales favorably with the lifetime of the quantum memory. Our work may have important applications in building distributed quantum networks through quantum memory-assisted atom-light interferometers.

quant-ph↗
arXiv2026

NNLO QCD corrections to hadron production in DIS at finite transverse momentum

We present the first calculation of hadron production in deep-inelastic scattering (DIS) at finite transverse momentum to next-to-next-to-leading order (NNLO) in perturbative QCD. To overcome the long-standing challenge of infrared divergences in semi-inclusive processes with identified final state hadrons at finite transverse momentum, we implement the recently developed qT-subtraction framework based on the recoil-free jet definition. By utilizing the winner-take-all recombination scheme, we achieve a consistent factorization for hadron-jet associated production, enabling the inclusion of $O(α_s^3)$ corrections. Our NNLO results generally demonstrate an improved convergence of the perturbative expansion and a reduction in scale uncertainties compared to previous next-to-leading order ones, especially for comparisons to multiplicity data from the ZEUS Collaboration. This work provides a high-precision theoretical foundation for the upcoming electron-ion collider era and establishes a new benchmark for the exploration of the nucleon's three-dimensional structure.

hep-ph↗
arXiv2026

Polarized Semi-Inclusive Deep-Inelastic Scattering at $\mathcal O(α_s^3)$ in QCD

Unraveling the partonic origin of the proton spin requires precise determinations of polarized parton distribution functions (PDFs), which depend on comparably precise theoretical predictions for polarized scattering, particularly in view of the high-precision measurements anticipated at the future Electron-Ion Collider. We present the first next-to-next-to-next-to-leading order (N$^3$LO) QCD predictions for longitudinally polarized semi-inclusive deep-inelastic scattering (SIDIS) in a fully differential form, together with next-to-next-to-leading order predictions for the hadron transverse-momentum spectrum. These results are obtained by extending the two-dimensional transverse-momentum subtraction framework to the spin-dependent cross section. Together with the corresponding unpolarized calculation, these results enable consistent N$^3$LO predictions for longitudinal double-spin asymmetries and provide a precision baseline for future analyses of helicity PDFs and transverse-momentum-dependent helicity distributions.

hep-ph↗
arXiv2026

Day-timescale Quasi-periodic Oscillations of the Gev BL Lac RX J0805.4+7534 with TESS

This paper reports for the first time the detection of quasi-periodic oscillations (QPOs) in the light curves of the BL Lacertae object RX J0805.4+7534. The Transiting Exoplanetary Survey Satellite (TESS) observed this source in seven sectors of the sky, and we extracted the light curves for these sectors using a custom method. The presence of QPO signals was found in these light curves. To detecte the periodicity and assess the statistical significance of the QPO signals, we employed two methods: Lomb-Scargle periodograms and Weighted Wavelet \textit{Z}-transform. Both of these different methods yielded consistent results. \textcolor{red}{\sout{The results showed signals with day-timescale QPO in sectors 20, 26, 53, and 73, with confidence levels exceeding $3σ$, and three of the sectors exhibited a ~3.8 days QPO. To explain these rapid quasi-periodic changes, we discuss several possible explanations. We propose that the QPO signal might be due to the rotation of hot spots on circular orbits within the accretion disk.}}\textcolor{blue}{The results show that QPO signals exist in sectors 20, 26, and 53, with the confidence levels exceeding 99.73\%. The QPOs in sectors 20 and 26 are $\sim 3.8$ days and have a global significance of 95\%. To explain these rapid quasi-periodic variations, we discussed several possible physical scenarios. The most possible one is the kink instability in relativistic jets. The other possible scenario is the rotation of hot-spots in the innermost accretion disk. Based on the hot-spot orbital model hypothesis, the mass of the black hole at the center of this BL Lac object wes estimated. The validity of these two explanations requires further observational data to verify. However, since the radiation of BL Lac objects primarily originates from jets, we prefer the kink instability in relativistic jets to be the cause of this rapid QPO.

astro-ph.HE↗
arXiv2026

SFMP: Fine-Grained, Hardware-Friendly and Search-Free Mixed-Precision Quantization for Large Language Models

Mixed-precision quantization is a promising approach for compressing large language models under tight memory budgets. However, existing mixed-precision methods typically suffer from one of two limitations: they either rely on expensive discrete optimization to determine precision allocation, or introduce hardware inefficiencies due to irregular memory layouts. We propose SFMP, a search-free and hardware-friendly mixed-precision quantization framework for large language models. The framework is built upon four novel ideas: Fractional bit-width, which extends integer bit-width for weight matrix to fractional value and transforms discrete precision allocation as a continuous problem; 2)Block-wise mixed-precision, enabling fine-grained precision within weight matrices while remaining hardware-friendly; 3)Row-column weight reordering, which aggregates salient weights via row and column reordering, incurring only a small activation reordering overhead during inference; 4)Unified GEMM kernel, which supports mixed-precision GEMM at arbitrary average bit-width. Extensive experiments demonstrate that SFMP outperforms state-of-the-art layer-wise mixed-precision methods under the same memory constraints, while significantly reducing quantization cost and improving inference efficiency. Code is available at https://github.com/Nkniexin/SFMP

cs.LG↗
arXiv2026

NeighborDiv: Training-free Zero-shot Generalist Graph Anomaly Detection via Neighbor Diversity

Graph Anomaly Detection (GAD) is increasingly shifting to Generalist GAD (GGAD) for cross-domain "one-for-all" detection, but existing GGAD methods predominantly rely on the neighbor consistency principle, falling into the \textbf{Node-to-Neighbor Consistency Paradigm} for anomaly quantification. These methods suffer from complex training pipelines, heavy training data dependency, high computational costs, and unstable cross-domain generalization. To address these limitations, we propose NeighborDiv, a training-free generalist graph anomaly detection framework based on neighbor diversity. Departing from the dominant Node-to-Neighbor Consistency Paradigm, we shift the focus to the \textbf{Neighbor-to-Neighbor Diversity Paradigm}, and uncover that the internal structural dispersion of a node's neighbor set is a powerful, independently discriminative anomaly signal. We quantify neighbor diversity via the variance of inter-neighbor feature similarities, which captures how a node organizes its local graph environment, and operates independently of conventional node-to-neighbor consistency frameworks. Extensive experiments under two standard GGAD evaluation paradigms show NeighborDiv achieves state-of-the-art performance, with relative gains of 10.25% in average AUC and 17.78% in average AP over the second-best baseline under Single-Domain Independent Training (SDIT), and 6.89%/9.58% in AUC/AP under Unified Multi-Domain Training (UMDT), respectively. Notably, NeighborDiv yields zero performance volatility across all datasets, eliminating training-set dependency and establishing a lightweight and highly practical GGAD framework.

cs.LG↗
arXiv2026

Hunyuan3D 2.0: Scaling Diffusion Models for High Resolution Textured 3D Assets Generation

We present Hunyuan3D 2.0, an advanced large-scale 3D synthesis system for generating high-resolution textured 3D assets. This system includes two foundation components: a large-scale shape generation model -- Hunyuan3D-DiT, and a large-scale texture synthesis model -- Hunyuan3D-Paint. The shape generative model, built on a scalable flow-based diffusion transformer, aims to create geometry that properly aligns with a given condition image, laying a solid foundation for downstream applications. The texture synthesis model, benefiting from strong geometric and diffusion priors, produces high-resolution and vibrant texture maps for either generated or hand-crafted meshes. Furthermore, we build Hunyuan3D-Studio -- a versatile, user-friendly production platform that simplifies the re-creation process of 3D assets. It allows both professional and amateur users to manipulate or even animate their meshes efficiently. We systematically evaluate our models, showing that Hunyuan3D 2.0 outperforms previous state-of-the-art models, including the open-source models and closed-source models in geometry details, condition alignment, texture quality, and etc. Hunyuan3D 2.0 is publicly released in order to fill the gaps in the open-source 3D community for large-scale foundation generative models. The code and pre-trained weights of our models are available at: https://github.com/Tencent/Hunyuan3D-2

cs.CV↗
arXiv2026

The Next-Generation 21CMA Telescope: Design, Commissioning, and Instrumental Effects in an SKA-LFAA-Like System

As the Square Kilometre Array (SKA) approaches operational status, its complex digital architecture introduces new instrumental challenges. To explore relevant observational and data processing strategies, we have upgraded the 21CMA telescope to the Next-Generation 21CMA (Ng21CMA). This paper presents the design and commissioning of the Ng21CMA system, featuring a digital backend capable of real-time beamforming. We demonstrate its performance through interferometric observations and high-time-resolution pulsar measurements, validating the system's sensitivity and operational stability. As a representative example of instrumental effects accessible with this platform, we investigate the impact of the two-stage channelization strategy used in SKA-LFAA-like systems. We show that it introduces a sawtooth-like spectral structure (SLOSS), characterized using both simulations and observational data. These results provide useful references for understanding instrument-induced spectral features and for guiding system design and calibration in future large-scale aperture arrays.

astro-ph.IM↗
arXiv2026

Two-Dimensional Transverse-Momentum Subtraction and Semi-Inclusive Deep-Inelastic Scattering at N$^3$LO in QCD

Identified hadron production is essential for the study of nucleon structure and QCD hadronization at high energies. We present the first calculation of unpolarized semi-inclusive deep-inelastic scattering (SIDIS) at next-to-next-to-next-to-leading order (N$^3$LO) in perturbative QCD. Our calculation is based on a novel method of two-dimensional transverse-momentum subtraction motivated by QCD factorization of soft and collinear singularities. The N$^3$LO corrections are moderate in general but can be significant in threshold regions, and exhibit excellent perturbative convergence and reduced scale variations. The fully differential framework allows for arbitrary selection cuts and directly enables precision nucleon tomography at the upcoming Electron-Ion Collider, establishing the theory foundation needed to match the anticipated experimental accuracy. Generalization of the method to calculations of polarized SIDIS is also feasible.

hep-ph↗
arXiv2026

QuarkMedBench: A Real-World Scenario Driven Benchmark for Evaluating Large Language Models

While Large Language Models (LLMs) excel on standardized medical exams, high scores often fail to translate to high-quality responses for real-world medical queries. Current evaluations rely heavily on multiple-choice questions, failing to capture the unstructured, ambiguous, and long-tail complexities inherent in genuine user inquiries. To bridge this gap, we introduce QuarkMedBench, an ecologically valid benchmark tailored for real-world medical LLM assessment. We compiled a massive dataset spanning Clinical Care, Wellness Health, and Professional Inquiry, comprising 20,821 single-turn queries and 3,853 multi-turn sessions. To objectively evaluate open-ended answers, we propose an automated scoring framework that integrates multi-model consensus with evidence-based retrieval to dynamically generate 220,617 fine-grained scoring rubrics (~9.8 per query). During evaluation, hierarchical weighting and safety constraints structurally quantify medical accuracy, key-point coverage, and risk interception, effectively mitigating the high costs and subjectivity of human grading. Experimental results demonstrate that the generated rubrics achieve a 91.8% concordance rate with clinical expert blind audits, establishing highly dependable medical reliability. Crucially, baseline evaluations on this benchmark reveal significant performance disparities among state-of-the-art models when navigating real-world clinical nuances, highlighting the limitations of conventional exam-based metrics. Ultimately, QuarkMedBench establishes a rigorous, reproducible yardstick for measuring LLM performance on complex health issues, while its framework inherently supports dynamic knowledge updates to prevent benchmark obsolescence.

cs.CL↗
arXiv2026

F10.7 Index Prediction: A Multiscale Decomposition Strategy with Wavelet Transform for Performance Optimization

In this study, we construct Dataset A for training, validation, and testing, and Dataset B to evaluate generalization. We propose a novel F10.7 index forecasting method using wavelet decomposition, which feeds F10.7 together with its decomposed approximate and detail signals into the iTransformer model. We also incorporate the International Sunspot Number (ISN) and its wavelet-decomposed signals to assess their influence on prediction performance. Our optimal method is then compared with the latest method from S. Yan et al. (2025) and three operational models (SWPC, BGS, CLS). Additionally, we transfer our method to the PatchTST model used in H. Ye et al. (2024) and compare our method with theirs on Dataset B. Key findings include: (1) The wavelet-based combination methods overall outperform the baseline using only F10.7 index. The prediction performance improves as higher-level approximate and detail signals are incrementally added. The Combination 6 method integrating F10.7 with its first to fifth level approximate and detail signals outperforms methods using only approximate or detail signals. (2) Incorporating ISN and its wavelet-decomposed signals does not enhance prediction performance. (3) The Combination 6 method significantly surpasses S. Yan et al. (2025) and three operational models, with RMSE, MAE, and MAPE reduced by 18.22%, 15.09%, and 8.57%, respectively, against the former method. It also excels across four different conditions of solar activity. (4) Our method demonstrates superior generalization and prediction capability over the method of H. Ye et al. (2024) across all forecast horizons. To our knowledge, this is the first application of wavelet decomposition in F10.7 prediction, substantially improving forecast performance.

astro-ph.IM↗
arXiv2026

Prediction of Major Solar Flares Using Interpretable Class-dependent Reward Framework with Active Region Magnetograms and Domain Knowledge

In this work, we develop, for the first time, a supervised classification framework with class-dependent rewards (CDR) to predict $\geq$MM flares within 24 hr. We construct multiple datasets, covering knowledge-informed features and line-of sight (LOS) magnetograms. We also apply three deep learning models (CNN, CNN-BiLSTM, and Transformer) and three CDR counterparts (CDR-CNN, CDR-CNN-BiLSTM, and CDR-Transformer). First, we analyze the importance of LOS magnetic field parameters with the Transformer, then compare its performance using LOS-only, vector-only, and combined magnetic field parameters. Second, we compare flare prediction performance based on CDR models versus deep learning counterparts. Third, we perform sensitivity analysis on reward engineering for CDR models. Fourth, we use the SHAP method for model interpretability. Finally, we conduct performance comparison between our models and NASA/CCMC. The main findings are: (1)Among LOS feature combinations, R_VALUE and AREA_ACR consistently yield the best results. (2)Transformer achieves better performance with combined LOS and vector magnetic field data than with either alone. (3)Models using knowledge-informed features outperform those using magnetograms. (4)While CNN and CNN-BiLSTM outperform their CDR counterparts on magnetograms, CDR-Transformer is slightly superior to its deep learning counterpart when using knowledge-informed features. Among all models, CDR-Transformer achieves the best performance. (5)The predictive performance of the CDR models is not overly sensitive to the reward choices.(6)Through SHAP analysis, the CDR model tends to regard TOTUSJH as more important, while the Transformer tends to prioritize R_VALUE more.(7)Under identical prediction time and active region (AR) number, the CDR-Transformer shows superior predictive capabilities compared to NASA/CCMC.

cs.LG↗