arXiv subjects

Junhao Zhu

Publications and source records attributed to Junhao Zhu.

At least 19 recordsLinked to original sources

arXiv2026

Neutralino dark matter in gauge mediation

We explore the potential of neutralino dark matter within the framework of gauge-mediated supersymmetry (SUSY) breaking. Gauge mediation offers an advantage over gravity mediation, as SUSY CP/flavor problems are avoided more easily, leading to significantly milder fine-tuning for electroweak symmetry-breaking compared to gravity mediation. In our models, the lightest neutralino, as the lightest SUSY particle (LSP), is a viable dark matter candidate, assuming a gravitino mass of $\mathcal{O}(100)~\mathrm{TeV}$. The models are formulated in five-dimensional space-time, where the SUSY breaking field and the matter fields are placed on separate branes to avoid issues related to flavor and CP violation, and the radion is stabilized by a supersymmetric Goldberger--Wise mechanism. Four distinct neutralino dark matter scenarios are studied: bino-wino coannihilation, higgsino dark matter, wino dark matter, and entropy-diluted bino dark matter. For each case, we determine the allowed parameter spaces and evaluate their consistency with existing experimental limits. Additionally, we examine the potential for testing these models through future investigations at the High-Luminosity Large Hadron Collider (HL-LHC) and through dark matter direct detection experiments.

hep-ph↗

arXiv2026

Learning to Be Guided: Shaping Latent Geometry for Classifier-Free Guidance

Classifier-free guidance (CFG) is a standard component of conditional diffusion models, and guided generation performance has become the \textit{de facto} measure of state-of-the-art models. Recent work on latent diffusion models seeks to learn latent spaces that are easier to model with diffusion. However, under CFG, generation additionally depends on how conditions are arranged in the latent space. We study this arrangement under discrete class conditions, termed \emph{class geometry}, and investigate what class geometry allows a latent space to work well with CFG. Under a Gaussian class model, we show that shifting variance from intra- to inter-class at fixed total variance makes classes more distinguishable under noise, and derive a noisy posterior objective whose per-sample descent direction is proportional to the oracle CFG correction. Based on the analysis, we propose Prototype Regularization (PR), a lightweight surrogate for the intractable noisy posterior objective, which applies a cosine prototype classifier to spatially pooled latents. With a controlled training budget (20 epochs for VAE, 80 epochs for DiT), adding PR to an LDM baseline reduces guided gFID on ImageNet from 3.31 to 2.28, surpassing a DINOv3-aligned VA-VAE (2.44) with negligible overhead. We further demonstrate that PR generalizes to pseudo-labels derived from clustering, continuous text-embedding conditions, and video latents. Together, our findings highlight class geometry as a critical factor in guided generation and take a first step toward CFG-compatible representation learning. Our project page is available at: https://maps-research.github.io/cfg-geometry/.

cs.LG↗

arXiv2026

Deep preferential evolution guided by pairwise comparisons enables motion correction in photoacoustic tomography

Motion artifacts in three-dimensional photoacoustic tomography (PAT), caused by extended mechanical scanning of sparse arrays, degrade image quality. Since sparse sampling results in low-quality sub-reconstructions, tracking motion directly is challenging. While final image quality could in principle guide artifact correction, conventional regularizers do not accurately reflect photoacoustic image quality. Here, we introduce deep preferential evolution (DPE), a derivative-free optimization framework where a learned image comparator guides evolutionary search over reconstruction parameters. By learning relative quality differences between same-target image pairs rather than absolute quality scores, the comparator generalized more robustly from simulation to in-vivo images (after unlabeled domain calibration) than an absolute scorer. Comparator-based DPE successfully corrected synthetic motion across diverse anatomies and backgrounds, and it mitigated artifacts from natural, unconstrained motion in human palm images. These results demonstrate that pairwise learned comparison can provide a transferable optimization objective for high-dimensional image restoration when labeled experimental training data are scarce.

physics.med-ph↗

arXiv2026

Coherence-Aware Distributional Evaluation of Open-Ended Text Generation

Existing open-ended generation metrics measure likelihood, lexical diversity, or distributional similarity in generic representation space, yet can miss fundamental dimensions of quality. A prominent blind spot is global coherence: a generated passage may be locally fluent while remaining globally contradictory, causally inconsistent, or topically disconnected. We identify representation as a central bottleneck in detecting these failures and introduce CHORD (Coherence-aware Hidden-state Open-generation Reference Distance), a coherence-sensitive distributional metric. CHORD encodes generated and human-written corpora in the hidden-state space of a frozen LLM using a coherence-eliciting prompt, and compares the resulting distributions using RBF-MMD. To test coherence sensitivity and selectivity, we construct a counterfactual evaluation suite pairing graded coherence-degrading perturbations with meaning-preserving controls. CHORD selectively detects relation, discourse, structural, and mixture failures that perplexity, entropy, MAUVE, FBD, and MMD-based baselines either miss or cannot separate from benign rewriting. Factorial ablations show that representation is the primary source of coherence sensitivity, while RBF-MMD improves sample efficiency. Larger backbones capture finer-grained distinctions, but coherence prompting improves selectivity only when the backbone can follow the prompt. On unconditional generation and prefix continuation, CHORD yields model rankings that strongly align with human judgments of whether outputs make sense and appear human-written. Together, these results establish representation design as central to reliable distributional evaluation. Code: https://github.com/MAPS-research/CHORD. Experiments: https://github.com/MAPS-research/CHORD-Experiment.

cs.CL↗

arXiv2026

FINSKILLOPS: A Self-Evolving Multi-Agent System for SEC Filing QA

Financial QA systems are typically improved before deployment through better retrieval, prompting, or agent coordination, leaving their reliability behavior fixed thereafter. In practice, new SEC-filing questions repeatedly expose heterogeneous errors in period, entity, evidence use, and calculation. Existing self-improvement methods can turn failures into new behaviors, but offer limited control over where a correction should apply or which previously correct answers it may break. We therefore frame post-deployment improvement as controlled behavioral maintenance: recurring failures should become scoped skill patches, and each patch should earn deployment with- out introducing regressions. We instantiate this view in FINSKILLOPS, a multi-agent system for SEC filing QA. FINSKILLOPS derives reusable skills from evidence-grounded, typed failure diagnoses and governs them through targeted validation, protected-case regression checks, negative controls, and versioned replacement or retirement. Across six financial QA benchmarks, a single frozen skill registry achieves the highest verdict-weighted correctness and reference consistency among the evaluated systems. Evolved skills raise correctness from 3.70 to 4.55 on our enhanced benchmark. In a separate 12-round operational study, only six of 33 proposed skills are promoted, while the monitoring non-correct rate falls from 20.0% to 12.5%. These results establish controlled skill scope, admission, and lifecycle management as the foundation for reliable self-improvement.

cs.AI↗

arXiv2026

QwenPaw-Data: Bridging Facts, Methodology, and Execution for Autonomous Enterprise Data Analytics

Enterprise data analysis is emerging as a distinct frontier for autonomous agents. Compared with general-purpose interaction and software engineering, it operates in an open, ambiguous, and continuously evolving environment. These characteristics call for a data-agent architecture that treats semantics, methodology, execution, and evolution as first-class system concerns. To this end, we introduce QwenPaw-Data, an agentic data system designed for enterprise intelligent data analysis. QwenPaw-Data consolidates heterogeneous assets from warehouses, dashboards, documents, interaction logs, and historical tasks into reusable, governable, and evolvable analysis assets, then turns natural-language requests into end-to-end analytical workflows spanning data understanding, retrieval, analysis, report generation, and decision support. Its architecture decomposes the problem into three collaborative subsystems: DataBridge provides trustworthy semantic grounding through interconnected metadata, knowledge, and trace graphs; Skill-Hub codifies expert analytical methodology into reusable and verifiable skills; and Host materializes these evidence and method assets into controllable, artifact-centric runtime execution. Across these subsystems, semantics, methods, traces, and feedback are continuously deposited back into the system, forming a self-evolving asset flywheel. Experiments on public benchmarks and real-world industrial BI workloads show that QwenPaw-Data improves both verifiable data access capability and higher-level analytical quality, offering a practical foundation for reliable, traceable, and continuously improving enterprise data agents.

cs.AI↗

arXiv2026

Replacing Multi-Step Assembly of Data Preparation Pipelines with One-Step LLM Pipeline Generation for Table QA

Table Question Answering (TQA) aims to answer natural language questions over structured tables. Large Language Models (LLMs) enable promising solutions to this problem, with operator-centric solutions that generate table manipulation pipelines in a multi-step manner offering state-of-the-art performance. However, these solutions rely on multiple LLM calls, resulting in prohibitive latencies and computational costs. We propose Operation-R1, the first framework that trains lightweight LLMs (e.g., Qwen-4B/1.7B) via a novel variant of reinforcement learning with verifiable rewards to produce high-quality data-preparation pipelines for TQA in a single inference step. To train such an LLM, we first introduce a self-supervised rewarding mechanism to automatically obtain fine-grained pipeline-wise supervision signals for LLM training. We also propose variance-aware group resampling to mitigate training instability. To further enhance robustness of pipeline generation, we develop two complementary mechanisms: operation merge, which filters spurious operations through multi-candidate consensus, and adaptive rollback, which offers runtime protection against information loss in data transformation. Experiments on two benchmark datasets show that, with the same LLM backbone, Operation-R1 achieves average absolute accuracy gains of 8.83 and 4.44 percentage points over multi-step preparation baselines, with 79\% table compression and a 2.2$\times$ reduction in monetary cost.

cs.DB↗

arXiv2026

Modeling Cell Developmental Trajectory using Multinomial Unbalanced Optimal Transport

Single-cell trajectory analysis aims to reconstruct the biological developmental processes of cells as they evolve over time, leveraging temporal correlations in gene expression. During cellular development, gene expression patterns typically change and vary across different cell types. A significant challenge in this analysis is that RNA sequencing destroys the cell, making it impossible to track gene expression across multiple stages for the same cell. Recent advances have introduced the use of optimal transport tools to model the trajectory of individual cells. In this paper, our focus shifts to a question of greater practical importance: we examine the differentiation of cell types over time. Specifically, we propose a novel method based on discrete unbalanced optimal transport to model the developmental trajectory of cell types. Our method detects biological changes in cell types and infers their transitions to different states by analyzing the transport matrix. We validated our method using single-cell RNA sequencing data from mouse embryonic fibroblasts. The results accurately identified major developmental changes in cell types, which were corroborated by experimental evidence. Furthermore, the inferred transition probabilities between cell types are highly congruent to biological ground truth.

stat.AP↗

arXiv2025

Beyond Relational: Semantic-Aware Multi-Modal Analytics with LLM-Native Query Optimization

Multi-modal analytical processing has the potential to transform applications in e-commerce, healthcare, entertainment, and beyond. However, real-world adoption remains elusive due to the limited ability of traditional relational query operators to capture query semantics. The emergence of foundation models, particularly the large language models (LLMs), opens up new opportunities to develop flexible, semantic-aware data analytics systems that transcend the relational paradigm. We present Nirvana, a multi-modal data analytics framework that incorporates programmable semantic operators while leveraging both logical and physical query optimization strategies, tailored for LLM-driven semantic query processing. Nirvana addresses two key challenges. First, it features an agentic logical optimizer that uses natural language-specified transformation rules and random-walk-based search to explore vast spaces of semantically equivalent query plans -- far beyond the capabilities of conventional optimizers. Second, it introduces a cost-aware physical optimizer that selects the most effective LLM backend for each operator using a novel improvement-score metric. To further enhance efficiency, Nirvana incorporates computation reuse and evaluation pushdown techniques guided by model capability hypotheses. Experimental evaluations on three real-world benchmarks demonstrate that Nirvana is able to reduce end-to-end runtime by 10%--85% and reduces system processing costs by 76% on average, outperforming state-of-the-art systems at both efficiency and scalability.

cs.DB↗

arXiv2025

Enhanced Quality Aware-Scalable Underwater Image Compression

Underwater imaging plays a pivotal role in marine exploration and ecological monitoring. However, it faces significant challenges of limited transmission bandwidth and severe distortion in the aquatic environment. In this work, to achieve the target of both underwater image compression and enhancement simultaneously, an enhanced quality-aware scalable underwater image compression framework is presented, which comprises a Base Layer (BL) and an Enhancement Layer (EL). In the BL, the underwater image is represented by controllable number of non-zero sparse coefficients for coding bits saving. Furthermore, the underwater image enhancement dictionary is derived with shared sparse coefficients to make reconstruction close to the enhanced version. In the EL, a dual-branch filter comprising rough filtering and detail refinement branches is designed to produce a pseudo-enhanced version for residual redundancy removal and to improve the quality of final reconstruction. Extensive experimental results demonstrate that the proposed scheme outperforms the state-of-the-art works under five large-scale underwater image datasets in terms of Underwater Image Quality Measure (UIQM).

eess.IV↗

arXiv2025

Evaluating Menu OCR and Translation: A Benchmark for Aligning Human and Automated Evaluations in Large Vision-Language Models

The rapid advancement of large vision-language models (LVLMs) has significantly propelled applications in document understanding, particularly in optical character recognition (OCR) and multilingual translation. However, current evaluations of LVLMs, like the widely used OCRBench, mainly focus on verifying the correctness of their short-text responses and long-text responses with simple layout, while the evaluation of their ability to understand long texts with complex layout design is highly significant but largely overlooked. In this paper, we propose Menu OCR and Translation Benchmark (MOTBench), a specialized evaluation framework emphasizing the pivotal role of menu translation in cross-cultural communication. MOTBench requires LVLMs to accurately recognize and translate each dish, along with its price and unit items on a menu, providing a comprehensive assessment of their visual understanding and language processing capabilities. Our benchmark is comprised of a collection of Chinese and English menus, characterized by intricate layouts, a variety of fonts, and culturally specific elements across different languages, along with precise human annotations. Experiments show that our automatic evaluation results are highly consistent with professional human evaluation. We evaluate a range of publicly available state-of-the-art LVLMs, and through analyzing their output to identify the strengths and weaknesses in their performance, offering valuable insights to guide future advancements in LVLM development. MOTBench is available at https://github.com/gitwzl/MOTBench.

cs.LG↗

arXiv2025

Five dimensional $SU(5) \times U(1)_{\rm PQ}$ grand unification

We propose a grand unified model in five dimensions that addresses the strong CP problem. In this framework, the boundary conditions that break the unified gauge group into the Standard Model gauge group simultaneously explain the mass splitting of the new matter content required for gauge coupling unification and the emergence of the QCD axion as the $U(1)_{\rm PQ}$ gauge field. The quality of the axion solution is inherently ensured by the gauge symmetry of the model. We further investigate the implications of this setup, including its predictions for the proton decay rate, the detectability of axion dark matter, and its cosmological impact. Notably, the model predicts a proton decay signal within the sensitivity of upcoming experiments, a result driven by the trade-off between the unification scale and the reheating temperature. In our scenario, the scales of grand unification and the axion decay constant are naturally unified.

hep-ph↗

arXiv2025

DIMT25@ICDAR2025: HW-TSC's End-to-End Document Image Machine Translation System Leveraging Large Vision-Language Model

This paper presents the technical solution proposed by Huawei Translation Service Center (HW-TSC) for the "End-to-End Document Image Machine Translation for Complex Layouts" competition at the 19th International Conference on Document Analysis and Recognition (DIMT25@ICDAR2025). Leveraging state-of-the-art open-source large vision-language model (LVLM), we introduce a training framework that combines multi-task learning with perceptual chain-of-thought to develop a comprehensive end-to-end document translation system. During the inference phase, we apply minimum Bayesian decoding and post-processing strategies to further enhance the system's translation capabilities. Our solution uniquely addresses both OCR-based and OCR-free document image translation tasks within a unified framework. This paper systematically details the training methods, inference strategies, LVLM base models, training data, experimental setups, and results, demonstrating an effective approach to document image machine translation.

cs.CV↗

arXiv2025

AIstorian lets AI be a historian: A KG-powered multi-agent system for accurate biography generation

Huawei has always been committed to exploring the AI application in historical research. Biography generation, as a specialized form of abstractive summarization, plays a crucial role in historical research but faces unique challenges that existing large language models (LLMs) struggle to address. These challenges include maintaining stylistic adherence to historical writing conventions, ensuring factual fidelity, and handling fragmented information across multiple documents. We present AIstorian, a novel end-to-end agentic system featured with a knowledge graph (KG)-powered retrieval-augmented generation (RAG) and anti-hallucination multi-agents. Specifically, AIstorian introduces an in-context learning based chunking strategy and a KG-based index for accurate and efficient reference retrieval. Meanwhile, AIstorian orchestrates multi-agents to conduct on-the-fly hallucination detection and error-type-aware correction. Additionally, to teach LLMs a certain language style, we finetune LLMs based on a two-step training approach combining data augmentation-enhanced supervised fine-tuning with stylistic preference optimization. Extensive experiments on a real-life historical Jinshi dataset demonstrate that AIstorian achieves a 3.8x improvement in factual accuracy and a 47.6% reduction in hallucination rate compared to existing baselines. The data and code are available at: https://github.com/ZJU-DAILY/AIstorian.

cs.CL↗

arXiv2025

De-aberration for transcranial photoacoustic computed tomography through an adult human skull

Noninvasive transcranial photoacoustic computed tomography (PACT) of the human brain, despite its clinical potential as a complementary technology to functional MRI, remains impeded by the acoustic distortion induced by the human skull. The distortion, which is attributed to the markedly different material properties of the skull relative to soft tissue, results in heavily aberrated PACT images -- a problem that has remained unsolved for the past two decades. Herein, we report the first successful experimental demonstration of the de-aberration of PACT images through an ex-vivo adult human skull using a homogeneous elastic model for the skull. Using only the geometry, position, and orientation of the skull, we faithfully de-aberrate the PACT images of light-absorbing phantoms acquired through an ex-vivo human skull for different levels of phantom complexity and positions. We also demonstrate the generality of our results by attaining a similar extent of de-aberration through a second ex-vivo human skull. Our work addresses the longstanding challenge of skull-induced aberrations in transcranial PACT and advances the field towards unlocking the full potential of transcranial human brain PACT.

physics.med-ph↗

arXiv2024

Stationarity of Manifold Time Series

In modern interdisciplinary research, manifold time series data have been garnering more attention. A critical question in analyzing such data is ``stationarity'', which reflects the underlying dynamic behavior and is crucial across various fields like cell biology, neuroscience and empirical finance. Yet, there has been an absence of a formal definition of stationarity that is tailored to manifold time series. This work bridges this gap by proposing the first definitions of first-order and second-order stationarity for manifold time series. Additionally, we develop novel statistical procedures to test the stationarity of manifold time series and study their asymptotic properties. Our methods account for the curved nature of manifolds, leading to a more intricate analysis than that in Euclidean space. The effectiveness of our methods is evaluated through numerical simulations and their practical merits are demonstrated through analyzing a cell-type proportion time series dataset from a paper recently published in Cell. The first-order stationarity test result aligns with the biological findings of this paper, while the second-order stationarity test provides numerical support for a critical assumption made therein.

stat.ME↗

arXiv2024

LLOT: application of Laplacian Linear Optimal Transport in spatial transcriptome reconstruction

Single-cell RNA sequencing (scRNA-seq) allows transcriptional profiling, and cell-type annotation of individual cells. However, sample preparation in typical scRNA-seq experiments often homogenizes the samples, thus spatial locations of individual cells are often lost. Although spatial transcriptomic techniques, such as in situ hybridization (ISH) or Slide-seq, can be used to measure gene expression in specific locations in samples, it remains a challenge to measure or infer expression level for every gene at a single-cell resolution in every location in tissues. Existing computational methods show promise in reconstructing these missing data by integrating scRNA-seq data with spatial expression data such as those obtained from spatial transcriptomics. Here we describe Laplacian Linear Optimal Transport (LLOT), an interpretable method to integrate single-cell and spatial transcriptomics data to reconstruct missing information at a whole-genome and single-cell resolution. LLOT iteratively corrects platform effects and employs Laplacian Optimal Transport to decompose each spot in spatial transcriptomics data into a spatially-smooth probabilistic mixture of single cells. We benchmarked LLOT against several other methods on datasets of Drosophila embryo, mouse cerebellum and synthetic datasets generated by scDesign3 in the paper, and another three datasets in the supplementary. The results showed that LLOT consistently outperformed others in reconstructing spatial expressions.

stat.AP↗

arXiv2024

CoachLM: Automatic Instruction Revisions Improve the Data Quality in LLM Instruction Tuning

Instruction tuning is crucial for enabling Language Learning Models (LLMs) in responding to human instructions. The quality of instruction pairs used for tuning greatly affects the performance of LLMs. However, the manual creation of high-quality instruction datasets is costly, leading to the adoption of automatic generation of instruction pairs by LLMs as a popular alternative. To ensure the high quality of LLM-generated instruction datasets, several approaches have been proposed. Nevertheless, existing methods either compromise dataset integrity by filtering a large proportion of samples, or are unsuitable for industrial applications. In this paper, instead of discarding low-quality samples, we propose CoachLM, a novel approach to enhance the quality of instruction datasets through automatic revisions on samples in the dataset. CoachLM is trained from the samples revised by human experts and significantly increases the proportion of high-quality samples in the dataset from 17.7% to 78.9%. The effectiveness of CoachLM is further assessed on various real-world instruction test sets. The results show that CoachLM improves the instruction-following capabilities of the instruction-tuned LLM by an average of 29.9%, which even surpasses larger LLMs with nearly twice the number of parameters. Furthermore, CoachLM is successfully deployed in a data management system for LLMs at Huawei, resulting in an efficiency improvement of up to 20% in the cleaning of 40k real-world instruction pairs. We release various assets of CoachLM, including the training data, code and test set (https://github.com/lunyiliu/CoachLM).

cs.CL↗