arXiv subjects

Xinyu Dai

Publications and source records attributed to Xinyu Dai.

At least 37 records · Page 2Linked to original sources

arXiv2026

Detection of Variability in Seyfert 2 Galaxies and Measurement of the Optical Scattering Region Size

One main theme of the Unification Model of active galactic nuclei is that there is an obscuring torus structure blocking the direct view of the central engine for Seyfert 2 galaxies. Here, we present the detection of long-term optical variability for a sample of nearby Seyfert 2s. We found that Seyfert 2s exhibit a relatively low but significant level of variability beyond the constant fluxes established by galaxies over month to year time scales. The variability is also detected in the structure functions of Seyfert 2s. Assuming a simple variability suppression model by the scattering region and dilution due to host starlight, where the region smooths the unobscured photon packets from the central engine as the light scatters over the torus, we estimate AGN scattering region sizes by matching the variability amplitudes of Seyfert 1s to 2s. Our measured scattering sizes are largely consistent with the torus size measured using their emission properties, suggesting that the scattering region is of a similar size as the torus. Our results pave the way towards variability as a powerful and independent test for AGN unification models.

astro-ph.GA↗

arXiv2026

GePBench: Evaluating Fundamental Geometric Perception for Multimodal Large Language Models

Geometric shapes play important roles in both physical world and human cognition. While multimodal large language models (MLLMs) have made significant advancements in visual understanding, their abilities to recognize geometric shapes and their spatial relationships, which we term \emph{geometric perception}, are not explicitly and systematically explored. To address this gap, we introduce GePBench, a novel benchmark specifically designed to assess the geometric perception capabilities of MLLMs. Our extensive evaluations reveal that even the current state-of-the-art MLLMs exhibit significant deficiencies in geometric perception tasks. Furthermore, we show that models trained with GePBench data demonstrate considerable improvements on a wide range of downstream tasks, highlighting the critical role of geometric perception in enabling advanced multimodal applications. Our code and datasets are available at \href{https://github.com/Changhao-Xiang/GePBench}{https://github.com/Changhao-Xiang/GePBench}.

cs.CL↗

arXiv2026

PaLMR: Towards Faithful Visual Reasoning via Multimodal Process Alignment

Reinforcement learning has recently improved the reasoning ability of Large Language Models and Multimodal LLMs, yet prevailing reward designs emphasise final-answer correctness and consequently tolerate process hallucinations--cases where models reach the right answer while misperceiving visual evidence. We address this process-level misalignment with PaLMR, a framework that aligns not only outcomes but also the reasoning process itself. PaLMR comprises two complementary components: a perception-aligned data layer that constructs process-aware reasoning data with structured pseudo-ground-truths and verifiable visual facts, and a process-aligned optimisation layer that constructs a hierarchical reward fusion scheme with a process-aware scoring function to encourage visually faithful chains-of-thought and improve training stability. Experiments on Qwen2.5-VL-7B show that our approach substantially reduces reasoning hallucinations and improves visual reasoning fidelity, achieving state-of-the-art results on HallusionBench while maintaining strong performance on MMMU, MathVista, and MathVerse. These findings indicate that PaLMR offers a principled and practical route to process-aligned multimodal reasoning, advancing the reliability and interpretability of MLLMs.

cs.CV↗

arXiv2026

Recognize Your Orchestrator: An Entropy Dynamics Perspective for LLM Multi-Agent Systems

The transition from single-turn models to Multi-Agent Systems (MAS) promises enhanced problem-solving capabilities, yet the centralized orchestration topology remains a critical point of fragility. To analyze this, we propose a Mean-Field Entropy Dynamics framework, modeling the orchestration process as a system governed by the competing forces of task resolution and cumulative context loading. To facilitate validation, we introduce Inverse Workflow Generation (IWG), a multi-agent pipeline that synthesizes process-verifiable, high-complexity benchmarks with dense intermediate checkpoints. We demonstrate that our entropy dynamics model fits empirical trajectories, providing physically interpretable parameters that quantify system stability and performance collapse. Crucially, our analysis uncovers a ``Reasoning Trap": while reasoning-heavy models excel in isolated tasks, they frequently fail as orchestrators due to context squeezing. Elucidating the physical mechanisms underlying the Orchestrator and quantifying systemic uncertainty offers insights for the MASs' architectural design.

cs.AI↗

arXiv2026

PEMark: Watermarking API Responses Based on Proxy Gateways and Position Encoding

Data leakage from API responses has drawn wide attention. APIs are often not fully regulated, making them easy to abuse. One common solution is to embed watermarks into API responses for traceability. However, existing watermarking methods often require modifying database content or API response data. This forces changes to business system code, and may even disrupt normal business operations because data values are altered. In this paper, we propose an original pluggable watermarking scheme based on a watermark proxy gateway and PEMark (Position Encoding-based Watermarking). The key novelty of our approach is exploiting the inherent permutation redundancy in the ordering of JSON/XML key-value pairs -- an overlooked dimension that carries no semantic information yet provides abundant encoding capacity. First, we forward server responses to the watermark proxy gateway, a design that requires zero modification to existing business systems. Then, we embed a watermark into each API response using position encoding, which reorders keys without altering any data values. To the best of our knowledge, this is the first work to achieve distortion-free API response watermarking via position encoding over a proxy gateway. Our method does not modify any data values, so normal business operations continue seamlessly after watermark embedding. Experimental results show that our framework maintains business usability while ensuring that returned API data is traceable. Compared with current mainstream schemes, our method is robust against tampering and insertion attacks (100\% similarity), and can withstand certain levels of deletion attacks.

cs.CR↗

arXiv2026

Causal Evidence for Attention Head Imbalance in Modality Conflict Hallucination

Modality-conflict hallucination occurs when multimodal large language models (MLLMs) prioritize erroneous textual premises over contradictory visual evidence. To understand why visual evidence fails to prevail during generation, we take a mechanistic perspective and examine which internal components drive or resist this failure. We perform head-level causal analysis using path patching across five open-source MLLMs and identify two groups of attention heads with opposing causal roles: hallucination-driving heads and hallucination-resisting heads. We find a consistent asymmetry: driving effects are more broadly distributed and carry greater aggregate weight, whereas resisting effects concentrate in a small number of high-importance heads. Ablation experiments further confirm that these groups exert opposing effects during generation: distributed driving influence and localized resistance together form an imbalanced routing structure that biases generation toward the erroneous premise. Motivated by this finding, we propose MACI (Modality-conflict-Aware Causal Intervention), a conditional intervention that suppresses causally identified hallucination-driving heads only when conflict is detected. Across five MLLMs, MACI achieves the largest hallucination reduction among compared inference-time baselines on the MMMC benchmark with a favorable hallucination-accuracy trade-off, and transfers zero-shot to the SCI-SemanticConflict test.

cs.AI↗

arXiv2026

CogGen: A Cognitively Inspired Recursive Framework for Deep Research Report Generation

The autonomous synthesis of deep research reports represents a critical frontier for Large Language Models (LLMs), demanding sophisticated information orchestration and non-linear narrative logic. Current approaches rely on rigid predefined linear workflows, which cause error accumulation, preclude global restructuring from subsequent insights, and ultimately limit in-depth multimodal fusion and report quality. We propose CogGen, a Cognitively inspired recursive framework for deep research report Generation. Leveraging a Hierarchical Recursive Architecture to simulate cognitive writing, CogGen enables flexible planning and global restructuring. To extend this recursivity to multimodal content, we introduce Abstract Visual Representation (AVR): a concise intent-driven language that iteratively refines visual-text layouts without pixel-level regeneration overhead. We further present CLEF, a Cognitive Load Evaluation Framework, and curate a new benchmark from Our World in Data (OWID). Extensive experiments show CogGen achieves state-of-the-art results among open-source systems, generating reports comparable to professional analysts' outputs and surpassing Gemini Deep Research. Our code and dataset are available at https://github.com/NJUNLP/CogGen.

cs.MA↗

arXiv2026

The shortest detected intra-day variability of active galactic nuclei in TESS survey

AGNs are known to be variable in almost all wavelengths and timescales. The shortest variability timescale of AGNs can be used to probe the smallest scale structures within AGNs. We aim to measure the shortest detected variability timescale, $t_{min,ul}$, of type 1 radio-quiet Seyfert galaxies and analyse their characteristics. We extracted TESS light curves of 47 Seyfert 1 galaxies. We measured the PSDs of the sample, modelled by a power law model plus a constant noise, and constrained the shortest detected AGN variability timescale as the power law component exceeds the constant noise and systematic uncertainties indicated by the upper limits of non-variable quiescent galaxies' PSDs. We measured the upper limits of the shortest variability timescale to be $\log(t_{min,ul}/hrs)=0.85\pm0.55$. We compared these upper limits to a range of theoretical AGN variability timescales, and the natural interpretation of our measured $t_{min,ul}$ is the light crossing scale from a coherently varying region, where the measured $t_{min,ul}$ corresponds to the range from a few to thousands of gravitational radii. A significant fraction of these light crossing scales is smaller than the accretion disk emission sizes measured by quasar microlensing, reverberation mapping, or theoretical accretion disk models. Since we only measure the upper limits, the true physical shortest variability timescales are even shorter. We measure the power law index to be $2.0\pm0.2$, and find weak anticorrelations with the black hole mass and luminosity. Our analysis suggests that the shortest optical variability is driven by a compact region smaller than the accretion disk size, potentially by X-ray reprocessing. Alternatively, this shortest timescale variability suggests that the accretion disk can be inhomogeneous potentially caused by turbulence from magnetorotational instability or magnetic reconnections. (abridged)

astro-ph.GA↗

arXiv2026

Lookahead Tree-Based Rollouts for Enhanced Trajectory-Level Exploration in Reinforcement Learning with Verifiable Rewards

Reinforcement Learning with Verifiable Rewards (RLVR), particularly with algorithms like Group Relative Policy Optimization (GRPO), has proven highly effective in enhancing the reasoning capabilities of large language models. However, a critical bottleneck in current pipelines lies in the limited diversity of sampled trajectories during group rollouts. Homogeneous trajectories and their associated rewards would diminish the return signals for policy updates, thereby hindering effective policy learning. This lack of diversity stems primarily from token-level stochastic sampling, where local variations are likely to collapse into near-identical reasoning paths. To address this limitation, we propose Lookahead Tree-Based Rollouts (LATR), a novel rollout strategy designed to explicitly promotes trajectory-level diversity by enforcing branching into different candidate tokens likely to yield distinct continuations. Specifically, LATR iteratively operates in three stages: (1) branching at high-uncertainty generation steps, (2) performing lookahead simulation for each new branch, and (3) pruning branches that exhibits prolonged similarity during simulation. Compared with stochastic Sampling, LATR accelerates policy learning by 131% on average and improves final pass@1 performance by 4.2% on both GRPO and Dynamic sAmpling Policy Optimization (DAPO) algorithms across different reasoning tasks. Our code and data are publicly available at https://github.com/starreeze/latr.

cs.CL↗

arXiv2026

A Decade-Long Increasing Mid-Infrared Luminosity in Galaxy NGC6447: a Turning-On Candidate of Active Galactic Nucleus

It is widely expected that the obscured accretion stage can be the initial turning-on stage of active galactic nuclei from quiescent galaxies. We present mid-infrared light curves of NGC 6447 in 3.5$μ$m and 4.6 $μ$m bands observed by WISE/NEOWISE, which show an almost monotonic increasing trend of 1.2 mag over 14 years. The optical light curve from ASAS-SN during the same period is consistent with a constant showing no variability. The mid-infrared color evolution shows that the galaxy transitioned into an active galactic nucleus (AGN) in 2018. The SPHEREx spectrum reveals an increasing continuum resembling warm to hot dust emission from an AGN. NuSTAR detected an X-ray source with a 2-30 keV luminosity of $8.4\times10^{41}$ ergs/s at the lower boundary of AGN X-ray emission range, and a factor of >7 variability in one year compared to the Swift upper limit. NGC 6447 was classified as a quiescent galaxy in the literature. The multi-wavelength timing and spectral properties of NGC 6447 are consistent with the expected AGN turning on event, where the obscuring material around the AGN central engine is gradually dispersed, revealing the central engine. This example shows that long-term infrared variability can be a powerful tool to find similar sources. Based on the sample selection statistics, we estimate the duration of the episodes of AGN accretion (duty cycle) signified by the turning-on event as $10^4$-$10^6$ yr.

astro-ph.GA↗

arXiv2025

Persona-Aware Alignment Framework for Personalized Dialogue Generation

Personalized dialogue generation aims to leverage persona profiles and dialogue history to generate persona-relevant and consistent responses. Mainstream models typically rely on token-level language model training with persona dialogue data, such as Next Token Prediction, to implicitly achieve personalization, making these methods tend to neglect the given personas and generate generic responses. To address this issue, we propose a novel Persona-Aware Alignment Framework (PAL), which directly treats persona alignment as the training objective of dialogue generation. Specifically, PAL employs a two-stage training method including Persona-aware Learning and Persona Alignment, equipped with an easy-to-use inference strategy Select then Generate, to improve persona sensitivity and generate more persona-relevant responses at the semantics level. Through extensive experiments, we demonstrate that our framework outperforms many state-of-the-art personalized dialogue methods and large language models.

cs.CL↗

arXiv2025

TrustJudge: Inconsistencies of LLM-as-a-Judge and How to Alleviate Them

The adoption of Large Language Models (LLMs) as automated evaluators (LLM-as-a-judge) has revealed critical inconsistencies in current evaluation frameworks. We identify two fundamental types of inconsistencies: (1) Score-Comparison Inconsistency, where lower-rated responses outperform higher-scored ones in pairwise comparisons, and (2) Pairwise Transitivity Inconsistency, manifested through circular preference chains (A>B>C>A) and equivalence contradictions (A=B=C\neq A). We argue that these issues come from information loss in discrete rating systems and ambiguous tie judgments during pairwise evaluation. We propose TrustJudge, a probabilistic framework that addresses these limitations through two key innovations: 1) distribution-sensitive scoring that computes continuous expectations from discrete rating probabilities, preserving information entropy for more precise scoring, and 2) likelihood-aware aggregation that resolves transitivity violations using bidirectional preference probabilities or perplexity. We also formalize the theoretical limitations of current LLM-as-a-judge frameworks and demonstrate how TrustJudge's components overcome them. When evaluated with Llama-3.1-70B-Instruct as judge using our dataset, TrustJudge reduces Score-Comparison inconsistency by 8.43% (from 23.32% to 14.89%) and Pairwise Transitivity inconsistency by 10.82% (from 15.22% to 4.40%), while maintaining higher evaluation accuracy. Our work provides the first systematic analysis of evaluation framework inconsistencies in LLM-as-a-judge paradigms, offering both theoretical insights and practical solutions for reliable automated assessment. The framework demonstrates consistent improvements across various model architectures and scales, enabling more trustworthy LLM evaluation without requiring additional training or human annotations. The codes can be found at https://github.com/TrustJudge/TrustJudge.

cs.AI↗

arXiv2025

RealBench: A Chinese Multi-image Understanding Benchmark Close to Real-world Scenarios

While various multimodal multi-image evaluation datasets have been emerged, but these datasets are primarily based on English, and there has yet to be a Chinese multi-image dataset. To fill this gap, we introduce RealBench, the first Chinese multimodal multi-image dataset, which contains 9393 samples and 69910 images. RealBench distinguishes itself by incorporating real user-generated content, ensuring high relevance to real-world applications. Additionally, the dataset covers a wide variety of scenes, image resolutions, and image structures, further increasing the difficulty of multi-image understanding. Ultimately, we conduct a comprehensive evaluation of RealBench using 21 multimodal LLMs of different sizes, including closed-source models that support multi-image inputs as well as open-source visual and video models. The experimental results indicate that even the most powerful closed-source models still face challenges when handling multi-image Chinese scenarios. Moreover, there remains a noticeable performance gap of around 71.8\% on average between open-source visual/video models and closed-source models. These results show that RealBench provides an important research foundation for further exploring multi-image understanding capabilities in the Chinese context.

cs.CL↗

arXiv2025

Formation of Cosmic Noon Protogalaxies via Quasar-Induced Fragmentation of a Cosmic Filament

When black hole jets encounter ambient medium, they can compress the gas, trigger star formation, and create stellar clusters containing tens of thousands of stars. Here, we report a remarkable discovery of such a phenomenon that happened just 2.2 billion years after the Big Bang, during the Cosmic Noon era. Quasar SDSSJ141924.44+532315.5, powered by a one-billion-solar-mass black hole, is seen blasting a powerful jet that interacts with a hypermassive gas reservoir, creating a fascinating, clumpy, arc-like structure spanning over 250 kiloparsecs in projected length, consisting of at least eight clumps. Each clump contains billions of stars, is as massive as the Milky Way, and exhibits extreme levels of star formation. We interpret these findings as fragmentation of a cosmic filament triggered by a jet overpressurized expanding cocoon, which leads to the birth of protogalaxies, a process observed at scales never seen before. We find that the physical conditions within the filament are favorable for a fragmentation scenario to occur. We also discuss the survivability and evolution of individual clumps in the context of unsolved galaxy formation theory problems.

astro-ph.GA↗

arXiv2025

Sloan Digital Sky Survey-V: Pioneering Panoptic Spectroscopy

The Sloan Digital Sky Survey-V (SDSS-V) is pioneering panoptic spectroscopy: it is the first all-sky, multi-epoch, optical-to-infrared spectroscopic survey. SDSS-V is mapping the sky with multi-object spectroscopy (MOS) at telescopes in both hemispheres (the 2.5-m Sloan Foundation Telescope at Apache Point Observatory and the 100-inch du Pont Telescope at Las Campanas Observatory), where 500 zonal robotic fiber positioners feed light from a wide-field focal plane to an optical (R$\sim 2000$, 500 fibers) and a near-infrared (R$\sim 22,000$, 300 fibers) spectrograph. In addition to these MOS capabilities, the survey is pioneering ultra wide-field ($\sim$ 4000~deg$^2$) integral field spectroscopy enabled by a new dedicated facility (LVM-I) at Las Campanas Observatory, where an integral field spectrograph (IFS) with 1801 lenslet-coupled fibers arranged in a 0.5 degree diameter hexagon feeds multiple R$\sim$4000 optical spectrographs that cover 3600-9800 angstroms. SDSS-V's hardware and multi-year survey strategy are designed to decode the chemo-dynamical history of the Milky Way Galaxy and tackle fundamental open issues in stellar physics in its Milky Way Mapper program, trace the growth physics of supermassive black holes in its Black Hole Mapper program, and understand the self-regulation mechanisms and the chemical enrichment of galactic ecosystems at the energy-injection scale in its Local Volume Mapper program. The survey is well-timed to multiply the scientific output from major all-sky space missions. The SDSS-V MOS programs began robotic operations in 2021; IFS observations began in 2023 with the completion of the LVM-I facility. SDSS-V builds upon decades of heritage of SDSS's pioneering advances in data analysis, collaboration spirit, infrastructure, and product deliverables in astronomy.

astro-ph.IM↗

arXiv2025

PersuasiveToM: A Benchmark for Evaluating Machine Theory of Mind in Persuasive Dialogues

The ability to understand and predict the mental states of oneself and others, known as the Theory of Mind (ToM), is crucial for effective social scenarios. Although recent studies have evaluated ToM in Large Language Models (LLMs), existing benchmarks focus on simplified settings (e.g., Sally-Anne-style tasks) and overlook the complexity of real-world social interactions. To mitigate this gap, we propose PersuasiveToM, a benchmark designed to evaluate the ToM abilities of LLMs in persuasive dialogues. Our framework contains two core tasks: ToM Reasoning, which tests tracking of evolving desires, beliefs, and intentions; and ToM Application, which assesses the use of inferred mental states to predict and evaluate persuasion strategies. Experiments across eight leading LLMs reveal that while models excel on multiple questions, they struggle with the tasks that need tracking the dynamics and shifts of mental states and understanding the mental states in the whole dialogue comprehensively. Our aim with PersuasiveToM is to allow an effective evaluation of the ToM reasoning ability of LLMs with more focus on complex psychological activities. Our code is available at https://github.com/Yu-Fangxu/PersuasiveToM.

cs.CL↗

arXiv2025

Dynamic Decision-Making under Model Misspecification

In this study, I investigate the dynamic decision problem with a finite parameter space when the functional form of conditional expected rewards is misspecified. Traditional algorithms, such as Thompson Sampling, guarantee neither an $O(e^{-T})$ rate of posterior parameter concentration nor an $O(T^{-1})$ rate of average regret. However, under mild conditions, we can still achieve an exponential convergence rate of the parameter to a pseudo truth set, an extension of the pseudo truth parameter concept introduced by White (1982). I further characterize the necessary conditions for the convergence of the expected posterior within this pseudo-truth set. Simulations demonstrate that while the maximum a posteriori (MAP) estimate of the parameters fails to converge under misspecification, the algorithm's average regret remains relatively robust compared to the correctly specified case. These findings suggest opportunities to design simple yet robust algorithms that achieve desirable outcomes even in the presence of model misspecifications.

econ.EM↗

arXiv2025

Environmental Dependence of X-Ray Emission From The Least Massive Galaxies

The low-mass end of low-mass galaxies is largely unexplored in AGN studies, but it is essential for extending our understanding of the black hole-galaxy coevolution. We surveyed the 3D-HST catalog and collected a sample of 546 dwarf galaxies with stellar masses log(M$_*$/$M_\odot$)$<$8.7, residing in the GOODS-South deep field. We then used the unprecedented depth of Chandra available in the GOODS-South field to search for AGN. We carefully investigated the factors that could play roles in the AGN detectability, such as Chandra's point-spread function and the redshift- and off-axis-dependent detection limits. We identified 16 X-ray sources that are likely associated with AGN activity. Next, we evaluated the environment density of each galaxy by computing tidal indices. We uncovered a dramatic impact of the environment on AGN triggering as dwarfs from high-density environments showed an AGN fraction of 22.5\%, while the median stellar mass of this subset of dwarfs is only log(M$_*$/$M_\odot$)=8.1. In contrast, the low-density environment dwarfs showed an AGN fraction of only 1.4\%, in line with typically reported values from the literature. This highlights the fact that massive central black holes are ubiquitous even at the lowest mass scales and demonstrates the importance of the environment in triggering black hole accretion, as well as the necessity for deep X-ray data and proper evaluation of the X-ray data quality. Alternatively, even if the detected X-ray sources are related to stellar mass accretors rather than AGN, the environmental dependence persists, signaling the impact of the environment on galaxy evolution and star formation processes at the lowest mass scales. Additionally, we stacked the X-ray images of non-detected galaxies from high- and low-density environments, revealing similar trends.

astro-ph.GA↗