Searcharxiv⌕ Search

arXiv subjects

Liang Zhang

Publications and source records attributed to Liang Zhang.

At least 73 records · Page 4Linked to original sources
arXiv2026

ConMeZO: Adaptive Descent-Direction Sampling for Gradient-Free Finetuning of Large Language Models

Zeroth-order or derivative-free optimization (MeZO) is an attractive strategy for finetuning large language models (LLMs) because it eliminates the memory overhead of backpropagation. However, it converges slowly due to the inherent curse of dimensionality when searching for descent directions in the high-dimensional parameter space of billion-scale LLMs. We propose ConMeZO, a novel zeroth-order optimizer that accelerates convergence by adaptive directional sampling. Instead of drawing the direction uniformly at random, ConMeZO restricts the sampling to a cone centered around a momentum estimate. This concentrates the search in directions where the true gradient is more likely to lie and thus reduces the effect of high dimensions. We prove that ConMeZO achieves the same worst-case convergence rate as MeZO. Empirically, when finetuning LLMs on natural language tasks, ConMeZO is up to 2X faster than MeZO while retaining the low-memory footprint of zeroth-order methods.

cs.LG↗
arXiv2026

HCRE: LLM-based Hierarchical Classification for Cross-Document Relation Extraction with a Prediction-then-Verification Strategy

Cross-document relation extraction (RE) aims to identify relations between the head and tail entities located in different documents. Existing approaches typically adopt the paradigm of ``\textit{Small Language Model (SLM) + Classifier}''. However, the limited language understanding ability of SLMs hinders further improvement of their performance. In this paper, we conduct a preliminary study to explore the performance of Large Language Models (LLMs) in cross-document RE. Despite their extensive parameters, our findings indicate that LLMs do not consistently surpass existing SLMs. Further analysis suggests that the underperformance is largely attributed to the challenges posed by the numerous predefined relations. To overcome this issue, we propose an LLM-based \underline{H}ierarchical \underline{C}lassification model for cross-document \underline{RE} (HCRE), which consists of two core components: 1) an LLM for relation prediction and 2) a \textit{hierarchical relation tree} derived from the predefined relation set. This tree enables the LLM to perform hierarchical classification, where the target relation is inferred level by level. Since the number of child nodes is much smaller than the size of the entire predefined relation set, the hierarchical relation tree significantly reduces the number of relation options that LLM needs to consider during inference. However, hierarchical classification introduces the risk of error propagation across levels. To mitigate this, we propose a \textit{prediction-then-verification} inference strategy that improves prediction reliability through multi-view verification at each level. Extensive experiments show that HCRE outperforms existing baselines, validating its effectiveness.

cs.CL↗
arXiv2026

From Implicit to Explicit: Token-Efficient Logical Supervision for Mathematical Reasoning in LLMs

Recent studies reveal that large language models (LLMs) exhibit limited logical reasoning abilities in mathematical problem-solving, instead often relying on pattern-matching and memorization. We systematically analyze this limitation, focusing on logical relationship understanding, which is a core capability underlying genuine logical reasoning, and reveal that errors related to this capability account for over 90\% of incorrect predictions, with Chain-of-Thought Supervised Fine-Tuning (CoT-SFT) failing to substantially reduce these errors. To address this bottleneck, we propose First-Step Logical Reasoning (FSLR), a lightweight training framework targeting logical relationship understanding. Our key insight is that the first planning step-identifying which variables to use and which operation to apply-encourages the model to derive logical relationships directly from the problem statement. By training models on this isolated step, FSLR provides explicit supervision for logical relationship understanding, unlike CoT-SFT which implicitly embeds such relationships within complete solution trajectories. Extensive experiments across multiple models and datasets demonstrate that FSLR consistently outperforms CoT-SFT under both in-distribution and out-of-distribution settings, with average improvements of 3.2\% and 4.6\%, respectively. Moreover, FSLR achieves 4-6x faster training and reduces training token consumption by over 80\%.

cs.CL↗
arXiv2026

Sketch and Text Synergy: Fusing Structural Contours and Descriptive Attributes for Fine-Grained Image Retrieval

Fine-grained image retrieval via hand-drawn sketches or textual descriptions remains a critical challenge due to inherent modality gaps. While hand-drawn sketches capture complex structural contours, they lack color and texture, which text effectively provides despite omitting spatial contours. Motivated by the complementary nature of these modalities, we propose the Sketch and Text Based Image Retrieval (STBIR) framework. By synergizing the rich color and texture cues from text with the structural outlines provided by sketches, STBIR achieves superior fine-grained retrieval performance. First, a curriculum learning driven robustness enhancement module is proposed to enhance the model's robustness when handling queries of varying quality. Second, we introduce a category-knowledge-based feature space optimization module, thereby significantly boosting the model's representational power. Finally, we design a multi-stage cross-modal feature alignment mechanism to effectively mitigate the challenges of cross modal feature alignment. Furthermore, we curate the fine-grained STBIR benchmark dataset to rigorously validate the efficacy of our proposed framework and to provide data support as a reference for subsequent related research. Extensive experiments demonstrate that the proposed STBIR framework significantly outperforms state of the art methods.

cs.CV↗
arXiv2026

Beyond One-Size-Fits-All: Adaptive Test-Time Augmentation for Sequential Recommendation

Test-time augmentation (TTA) has become a promising approach for mitigating data sparsity in sequential recommendation by improving inference accuracy without requiring costly model retraining. However, existing TTA methods typically rely on uniform, user-agnostic augmentation strategies. We show that this "one-size-fits-all" design is inherently suboptimal, as it neglects substantial behavioral heterogeneity across users, and empirically demonstrate that the optimal augmentation operators vary significantly across user sequences with different characteristics for the first time. To address this limitation, we propose AdaTTA, a plug-and-play reinforcement learning-based adaptive inference framework that learns to select sequence-specific augmentation operators on a per-sequence basis. We formulate augmentation selection as a Markov Decision Process and introduce an Actor-Critic policy network with hybrid state representations and a joint macro-rank reward design to dynamically determine the optimal operator for each input user sequence. Extensive experiments on four real-world datasets and two recommendation backbones demonstrate that AdaTTA consistently outperforms the best fixed-strategy baselines, achieving up to 26.31% relative improvement on the Home dataset while incurring only moderate computational overhead

cs.IR↗
arXiv2026

Is Mathematical Problem-Solving Expertise in Large Language Models Associated with Assessment Performance?

Large Language Models (LLMs) are increasingly used in math education not only as problem solvers but also as assessors of learners' reasoning. However, it remains unclear whether stronger math problem-solving ability is associated with stronger step-level assessment performance. This study examines that relationship using the GSM8K and MATH subsets of PROCESSBENCH, a human-annotated benchmark for identifying the earliest erroneous step in mathematical reasoning. We evaluate two LLM-based math tutor agent settings, instantiated with GPT-4 and GPT-5, in two independent tasks on the same math problems: solving the original problem and assessing a benchmark-provided solution by predicting the earliest erroneous step. Results show a consistent within-model pattern: assessment accuracy is substantially higher on math problem items the same model solved correctly than on items it solved incorrectly, with statistically significant associations across both models and datasets. At the same time, assessment remains more difficult than direct problem solving, especially on error-present solutions. These findings suggest that math problem-solving expertise supports stronger assessment performance, but reliable step-level diagnosis also requires additional capabilities such as step tracking, monitoring, and precise error localization. The results have implications for the design and evaluation of AI-supported Adaptive Instructional Systems (AISs) for formative assessment in math education.

cs.AI↗
arXiv2026

GRMLR: Knowledge-Enhanced Small-Data Learning for Deep-Sea Cold Seep Stage Inference

Deep-sea cold seep stage assessment has traditionally relied on costly, high-risk manned submersible operations and visual surveys of macrofauna. Although microbial communities provide a promising and more cost-effective alternative, reliable inference remains challenging because the available deep-sea dataset is extremely small ($n = 13$) relative to the microbial feature dimension ($p = 26$), making purely data-driven models highly prone to overfitting. To address this, we propose a knowledge-enhanced classification framework that incorporates an ecological knowledge graph as a structural prior. By fusing macro-microbe coupling and microbial co-occurrence patterns, the framework internalizes established ecological logic into a \underline{\textbf{G}}raph-\underline{\textbf{R}}egularized \underline{\textbf{M}}ultinomial \underline{\textbf{L}}ogistic \underline{\textbf{R}}egression (GRMLR) model, effectively constraining the feature space through a manifold penalty to ensure biologically consistent classification. Importantly, the framework removes the need for macrofauna observations at inference time: macro-microbe associations are used only to guide training, whereas prediction relies solely on microbial abundance profiles. Experimental results demonstrate that our approach significantly outperforms standard baselines, highlighting its potential as a robust and scalable framework for deep-sea ecological assessment.

cs.LG↗
arXiv2026

MI-DPG: Decomposable Parameter Generation Network Based on Mutual Information for Multi-Scenario Recommendation

Conversion rate (CVR) prediction models play a vital role in recommendation and advertising systems. Recent research on multi-scenario recommendation shows that learning a unified model to serve multiple scenarios is effective for improving overall performance. However, it remains challenging to improve model prediction performance across scenarios at low model parameter cost, and current solutions are hard to robustly model multi-scenario diversity. In this paper, we propose MI-DPG for the multi-scenario CVR prediction, which learns scenario-conditioned dynamic model parameters for each scenario in a more efficient and effective manner. Specifically, we introduce an auxiliary network to generate scenario-conditioned dynamic weighting matrices, which are obtained by combining decomposed scenario-specific and scenario-shared low-rank matrices with parameter efficiency. For each scene, weighting the backbone model parameters by the weighting matrix helps to specialize the model parameters for different scenarios. It can not only modulate the complete parameter space of the backbone model but also improve the model effectiveness. Furthermore, we design a mutual information regularization to enhance the diversity of model parameters across different scenarios by maximizing the mutual information between the scenario-aware input and the scene-conditioned dynamic weighting matrix. Experiments from three real-world datasets show that MI-DPG significantly outperforms previous multi-scenario recommendation models.

cs.IR↗
arXiv2026

Coevolutionary dynamics of cooperation, risk, and cost in collective risk games

Addressing both natural and societal challenges requires collective cooperation. Studies on collective-risk social dilemmas have shown that individual decisions are influenced by the perceived risk of collective failure. However, existing feedback evolving game models often focus on a single feedback mechanism, such as the coupling between cooperation and risk or between cooperation and cost. In many real-world scenarios, however, the level of cooperation, the cost of cooperating, and the collective risk are dynamically interlinked. Here, we present an evolutionary game model that considers the interplay of these three variables. Our analysis shows that the worst-case scenario, characterized by full defection, maximum risk, and the highest cost of cooperation, remains a stable evolutionary attractor. Nevertheless, cooperation can emerge and persist because the system also supports stable equilibria with non-zero cooperation. The system exhibits multistability, meaning that different initial conditions lead to either sustained cooperation or a tragedy of the commons. These findings highlight that initial levels of cooperation, cost, and risk collectively determine whether a population can avert a tragic outcome.

nlin.AO↗
arXiv2026

Mocap-2-to-3: Multi-view Lifting for Monocular Motion Recovery with 2D Pretraining

Human motion recovery for real-world interaction demands both precise action details and metric-scale trajectories. Recovering absolute human pose from monocular input presents a viable solution, but faces two main challenges: (1) models' reliance on 3D training data from constrained environments limits their out-of-distribution generalization; and (2) the inherent difficulty of estimating metric-scale poses from monocular observations. This paper introduces Mocap-2-to-3, a novel framework that differs from prior HMR methods by recovering absolute poses from monocular input and leveraging abundant 2D data to enhance 3D motion recovery. To effectively utilize the action priors and diversity in large-scale 2D datasets, we reformulate 3D motion as a multi-view synthesis process and divide the training into two stages: a single-view diffusion model is first pre-trained on extensive 2D data, followed by multi-view fine-tuning on 3D data, thus achieving a combination of strong priors and geometric constraints. Furthermore, to recover absolute poses, we introduce a novel human motion representation that decouples the learning of local pose and global movements, while encoding ground geometric priors to accelerate convergence, thereby yielding more precise positioning in the physical world. Experiments on in-the-wild benchmarks show that our method outperforms state-of-the-art approaches in both camera-space motion realism and world-grounded human positioning, while exhibiting strong generalization capability.

cs.CV↗
arXiv2026

Lattice QCD study on nucleon-$Ω_{\rm ccc}$ interaction at the physical point

We report the S-wave interactions between the nucleon ($N$) and the triply charmed Omega baryon ($Ω_{\mathrm{ccc}}$) using (2+1)-flavor lattice QCD with a physical pion mass ($m_π\simeq 137.1$ MeV) on a lattice volume $\simeq (8.1~\mathrm{fm})^3$. The charm quark is implemented with a relativistic heavy-quark action at its physical mass. Employing the time-dependent HAL QCD method, the $N-Ω_{\mathrm{ccc}}$ potentials in the spin-1 ($^3\mathrm{S}_1$) and spin-2 ($^5\mathrm{S}_2$) channels are extracted. In both channels, overall attraction is found with the scattering parameters, $a_0 = 0.56(0.13)\left(^{+0.26}_{-0.03}\right)$ fm and $r_{\mathrm{eff}} = 1.60(0.05)\left(^{+0.04}_{-0.12}\right)$ fm for the $^3\mathrm{S}_1$ channel, and $a_0 = 0.38(0.12)\left(^{+0.25}_{-0.00}\right)$ fm and $r_{\mathrm{eff}} = 2.04(0.10)\left(^{+0.03}_{-0.22}\right)$ fm for the $^5\mathrm{S}_2$ channel, indicating the absence of a dibaryon bound state. The extracted potentials are further decomposed into spin-independent and spin-dependent components, which provides a useful handle to investigate the underlying interaction mechanism. The spin-independent potential is a dominant component and features a short-range attractive core and a long-range attractive tail, while the spin-dependent potential shows short-range attraction (repulsion) in the spin-1 (spin-2) channel. Qualitative comparisons with previous studies of the $N$-$J/ψ$ and $N-Ω_{\rm{sss}}$ systems at $m_π\simeq 146$ MeV are provided, emphasizing the role of heavy-hadron chromo-polarizability arising from soft-gluon exchange between the nucleon and flavor-singlet hadrons.

hep-lat↗
arXiv2026

A New Method for Identifying Contaminating Sources and Locating Target Sources through the Cross-Arm Features of Micro Pore Optics

The Pathfinder of the Type-A satellites in the Chasing All Transients Constellation Hunters (CATCH) space mission is equipped with Micro-Pore Optics (MPOs) and four single-pixel Silicon Drift Detectors (SDDs). Due to the lack of position resolution in an individual SDD, we propose a new method based on the cross-arms in the point spread function (PSF) of MPOs to enhance the satellite's capability in identifying contaminating sources and locating target sources. By placing one detector on each of the horizontal and vertical cross-arms on the focal plane, we can use the changes in the relative counts on the cross-arms detectors to deduce the location of the source. Simulated observations demonstrate that, for a target source with a flux of 1 Crab and an exposure time of 200 s, the cross-arms detectors can identify contaminating source with the same flux level at an off-axis angle larger than 8', and improve positioning accuracy to 6'. Furthermore, we extend the simulation study to CATCH Type-A, which plans to use an SDD array. In situations where sources exhibit the same flux of 1 Crab and the exposure time is merely 1 s, a 16x16 SDD array is capable of identifying contaminating source with an off-axis angle greater than 2.4' and can achieve a positioning precision of 1.8'.

astro-ph.IM↗
arXiv2026

Breaking Smooth-Motion Assumptions: A UAV Benchmark for Multi-Object Tracking in Complex and Adverse Conditions

The rapid movements and agile maneuvers of unmanned aerial vehicles (UAVs) induce significant observational challenges for multi-object tracking (MOT). However, existing UAV-perspective MOT benchmarks often lack these complexities, featuring predominantly predictable camera dynamics and linear motion patterns. To address this gap, we introduce DynUAV, a new benchmark for dynamic UAV-perspective MOT, characterized by intense ego-motion and the resulting complex apparent trajectories. The benchmark comprises 42 video sequences with over 1.7 million bounding box annotations, covering vehicles, pedestrians, and specialized industrial categories such as excavators, bulldozers and cranes. Compared to existing benchmarks, DynUAV introduces substantial challenges arising from ego-motion, including drastic scale changes and viewpoint changes, as well as motion blur. Comprehensive evaluations of state-of-the-art trackers on DynUAV reveal their limitations, particularly in managing the intertwined challenges of detection and association under such dynamic conditions, thereby establishing DynUAV as a rigorous benchmark. We anticipate that DynUAV will serve as a demanding testbed to spur progress in real-world UAV-perspective MOT, and we will make all resources available at link.

cs.CV↗
arXiv2026

Observationally Informed Adaptive Causal Experimental Design

Randomized Controlled Trials (RCTs) represent the gold standard for causal inference yet remain a scarce resource. While large-scale observational data is often available, it is utilized only for retrospective fusion, and remains discarded in prospective trial design due to bias concerns. We argue this "tabula rasa" data acquisition strategy is fundamentally inefficient. In this work, we propose Active Residual Learning, a new paradigm that leverages the observational model as a foundational prior. This approach shifts the experimental focus from learning target causal quantities from scratch to efficiently estimating the residuals required to correct observational bias. To operationalize this, we introduce the R-Design framework. Theoretically, we establish two key advantages: (1) a structural efficiency gap, proving that estimating smooth residual contrasts admits strictly faster convergence rates than reconstructing full outcomes; and (2) information efficiency, where we quantify the redundancy in standard parameter-based acquisition (e.g., BALD), demonstrating that such baselines waste budget on task-irrelevant nuisance uncertainty. We propose R-EPIG (Residual Expected Predictive Information Gain), a unified criterion that directly targets the causal estimand, minimizing residual uncertainty for estimation or clarifying decision boundaries for policy. Experiments on synthetic and semi-synthetic benchmarks demonstrate that R-Design significantly outperforms baselines, confirming that repairing a biased model is far more efficient than learning one from scratch.

stat.ML↗
arXiv2026

Super Research: Answering Highly Complex Questions with Large Language Models through Super Deep and Super Wide Research

While Large Language Models (LLMs) have demonstrated proficiency in Deep Research or Wide Search, their capacity to solve highly complex questions-those requiring long-horizon planning, massive evidence gathering, and synthesis across heterogeneous sources-remains largely unexplored. We introduce Super Research, a task for complex autonomous research tasks that integrates (i) structured decomposition into a research plan, (ii) super wide retrieval for diverse perspectives, and (iii) super deep investigation to resolve uncertainties through iterative queries. To evaluate this capability, we curated a benchmark of 300 expert-written questions across diverse domains, each requiring up to 100+ retrieval steps and 1,000+ web pages to reconcile conflicting evidence. Super Research produces verifiable reports with fine-grained citations and intermediate artifacts (e.g., outlines and tables) to ensure traceable reasoning. Furthermore, we present a graph-anchored auditing protocol that evaluates Super Research along five dimensions: Coverage, Logical Consistency, Report Utility, Objectivity and Citation Health. While super-complex questions may be infrequent in standard applications, Super Research serves as a critical ceiling evaluation and stress test for LLM capabilities. A model's proficiency within Super Research acts as a powerful proxy for its general research competence; success here suggests the robustness necessary to navigate nearly any subordinate research task. Leaderboard is available at: https://cnsdqd-dyb.github.io/Super-Research-Benchmark/

cs.CL↗
arXiv2026

Discovery of energy-dependent phase variations in the polarization angle of Cen X-3

We present a detailed polarimetric analysis of Cen X-3 using \ixpe observations during its high state, revealing complex energy-dependent polarization behavior. While phase-averaged polarization shows marginal energy dependence, phase-resolved analysis reveals that the energy dependence of the polarization angle (PA) is strongly phase-dependent, with dramatic variations visible in a few specific phase intervals. We model this behavior using a two-component polarization framework consisting of a pulsed component governed by the Rotating Vector Model (RVM) and an additional phase-dependent component. By allowing the additional component's polarized flux to vary with pulse phase while fixing its PA, the observed complex behavior can be reconciled with a single set of RVM parameters across all energies. Spectroscopic analysis using \ixpe, \nicer and \nustar during the high state reveals phase-modulated intrinsic hydrogen column density and covering fraction, suggesting that the wind properties are modulated with pulse phase. Our findings indicate that phase-dependent scattering in the disk wind may significantly alter the observed polarization properties of X-ray pulsars.

astro-ph.HE↗
arXiv2026

BEMEval-Doc2Schema: Benchmarking Large Language Models for Structured Data Extraction in Building Energy Modeling

Recent advances in foundation models, including large language models (LLMs), have created new opportunities to automate building energy modeling (BEM). However, systematic evaluation has remained challenging due to the absence of publicly available, task-specific datasets and standardized performance metrics. We present BEMEval, a benchmark framework designed to assess foundation models' performance across BEM tasks. The first benchmark in this suite, BEMEval-Doc2Schema, focuses on structured data extraction from building documentation, a foundational step toward automated BEM processes. BEMEval-Doc2Schema introduces the Key-Value Overlap Rate (KVOR), a metric that quantifies the alignment between LLM-generated structured outputs and ground-truth schema references. Using this framework, we evaluate two leading models (GPT-5 and Gemini 2.5) under zero-shot and few-shot prompting strategies across three datasets: HERS L100, NREL iUnit, and NIST NZERTF. Results show that Gemini 2.5 consistently outperforms GPT-5, and that few-shot prompts improve accuracy for both models. Performance also varies by schema: the EPC schema yields significantly higher KVOR scores than HPXML, reflecting its simpler and reduced hierarchical depth. By combining curated datasets, reproducible metrics, and cross-model comparisons, BEMEval-Doc2Schema establishes the first community-driven benchmark for evaluating LLMs in performing building energy modeling tasks, laying the groundwork for future research on AI-assisted BEM workflows.

cs.CE↗
arXiv2026

Can Wearable Exoskeletons Reduce Gender and Disability Gaps in the Construction Industry?

The share of construction trade jobs held by women and people with disabilities has remained stubbornly low in the face of chronic shortages of skilled labor. This study explores the potential of wearable assistive technologies to reduce these disparities. We use U.S. worker-level data to estimate employment and wage differences by gender and by mobility/strength impairments in construction and non-construction jobs. We also use occupational-level data to examine variations in workforce composition, physical skill requirements, and earnings across detailed construction occupations. Regression estimates indicate that being a woman and having strength and mobility impairments are associated with substantial employment and pay gaps in construction compared to non-construction jobs. Further analysis shows a high negative correlation between the representation of women and the ability levels required in those occupations. Finally, we discuss several wearable exoskeletons under development for people with upper-body and lower-body impairments, focusing on how these innovations could be integrated into construction jobs. These findings suggest that wearable exoskeletons that enhance manual dexterity, balance, and strength may improve the representation of women and people with disabilities in some of the higher-paying occupations in construction.

econ.GN↗