Searcharxiv⌕ Search

arXiv subjects

Kaixiang Wang

Publications and source records attributed to Kaixiang Wang.

At least 19 recordsLinked to original sources

S1-Evo: Experience-Guided Evidence Competition Attacks on Typed System-1 Decision Models

Typed System-1 (S1) decision models are increasingly deployed in LLM agents for routing, moderation, and authorization to cut inference cost, yet untrusted context can still manipulate typed outputs and alter execution branches. Unlike autoregressive System-2 (S2) models, S1 interfaces score dynamic rules and options without token-by-token decoding, mismatching attacks targeting text continuation or explicit override commands. We propose S1-Evo, a self-evolving framework tailored to S1's contract-based option competition and margin feedback. S1-Evo perturbs untrusted fields via contract-aware evidence channels to induce \emph{evidence competition} with original task evidence while preserving fact invariance. Offline on seed tasks, it optimizes candidates via feedback search guided by directional margins and re-verification, distilling verified flips into experience cards. At test time, S1-Evo retrieves relevant cards to construct perturbations for unseen tasks. Across four benchmarks and three S1 models, S1-Evo reaches 71.99\% mean ASR and 57.47\% high-confidence ASR on \texttt{jev-1.14}, and causes up to 24.00\% E2E agent violations.

cs.AI↗

ORCA:Orthogonal Residual Consensus Alignment for Multi-View Clustering

Deep multi-view clustering often suffers from redundant information accumulation during fusion and indiscriminate alignment across representation levels. We propose Orthogonal Residual Consensus Alignment (ORCA) to address these challenges. Its semantic incremental recurrent fusion module extracts cross-view orthogonal residuals as non-redundant increments and adaptively weights recurrent states through divergence-guided gating. Its consensus-calibrated hierarchical contrastive alignment module uses the global consensus to weight contrastive alignment of view-level, pairwise, and recurrent representations. Together, these modules balance view-specific structure with cross-view consistency. Experiments on five benchmarks against fifteen competing methods demonstrate ORCA's effectiveness, including a 5.25 percent accuracy improvement over the strongest baseline on 100leaves. Ablation studies validate both modules. Code is available at https://github.com/XiaojianDing/2026-NeurIPS-ORCA.

cs.CV↗

WARP-JEPA: World-Model-in-the-Loop Amortized Rollout Planning for Frozen Hierarchical Joint-Embedding Predictive Architectures

Hierarchical Joint-Embedding Predictive Architectures (H-JEPA) enable long-horizon visual control but rely on costly test-time gradient descent for planning. Our analysis reveals that iterative optimization consumes up to 97.7% of episode runtime and that low predicted rollout costs do not necessarily translate into successful control. We further identify a planning-horizon asymmetry between long-horizon top-level composition and short two-step subordinate planning. To address these challenges, we propose WARP-JEPA, a hierarchical amortized planning framework that replaces online gradient optimization with learned multi-candidate trajectory proposers while keeping the world models frozen. WARP-JEPA combines a lightweight world-model-in-the-loop causal Transformer at the top level with parallel residual MLPs at subordinate levels. The proposers are trained using model-based rollout costs and best-of-K behavior cloning, enabling efficient candidate selection without test-time gradient optimization. Across 30 frozen H-JEPA models on four benchmarks, WARP-JEPA achieves 9.6-18.8x faster episode inference on long-horizon tasks (2.9-3.0x on Cube) and improves task success by up to 33.3 percentage points. These results demonstrate that amortized planning can substantially improve both inference efficiency and control performance without retraining the world models. Code is available at https://github.com/hideinsewer/WARP-JEPA.

cs.AI↗

The Next Generation Virgo Cluster Survey (NGVS). XXIX. Demography of Nuclear Star Clusters

Nuclear star clusters (NSCs) provide insights into processes shaping galaxy cores. We use deep $u^*giz$ CFHT imaging from the Next Generation Virgo Cluster Survey (NGVS) to study NSCs in Virgo cluster galaxies. Our parent sample consists of 3680 certain, likely or possible cluster members spanning more than six orders of magnitude in stellar mass. Using a self-consistent classification framework, we identify 1144 candidate nucleated galaxies: 804 high-confidence NSCs and 340 lower-confidence candidates. This is the largest and most uniform NSC catalog for any environment studied to date. We explore the nucleation fraction, $f_{\rm nuc}$, as a function of stellar mass, morphology and environment. Consistent with previous studies, we find stellar mass a key factor driving nucleation probability. However, host-galaxy morphology is also important, with the peak of the $f_{\rm nuc}$ distribution occurring at progressively lower masses as one moves from later to earlier morphological types. We interpret this shift as evidence that the more extended structures of late-type galaxies reduce the efficiency of GC inspiral at fixed stellar mass, while gas-rich secular processes and central star formation may contribute increasingly to NSC growth at higher masses. Environment plays a smaller role, with only a slight enhancement in high-density regions. We characterize the nonlinear relationship between NSC stellar mass and host-galaxy mass, and assess the role of selection effects. We conclude that NSC formation is best explained by a hybrid scenario, with inspiraling star clusters and in situ star formation contributing, and their relative importance dictated primarily by galaxy mass.

astro-ph.GA↗

EGIP: An Evidence-Grounded Interaction Protocol for Clinical Agents

Interactive clinical agents acquire evidence through patient dialogue and diagnostic examinations, yet their performance also depends on how the interaction environment exposes information, available actions, and examination results. When these conditions are implicit or inconsistently controlled, final-answer accuracy can conflate diagnostic reasoning with the effects of the interaction protocol. We propose the \textbf{E}vidence-\textbf{G}rounded \textbf{I}nteraction \textbf{P}rotocol (EGIP), which enforces role-specific information boundaries, exposes diagnostic actions without revealing their results, retrieves examination results from stored clinical records, and organizes acquired evidence for final decision-making. By separating evidence acquisition from terminal prediction, EGIP enables different decision strategies to operate on the same observations. We evaluate EGIP across two clinical-agent frameworks, five datasets, and six language models. EGIP improves diagnostic accuracy over native baselines across all primary comparisons. A clinician review further shows that the proportion of automatically scored correct responses jointly judged evidence-supported rises from 10.20\% to 36.55\%. These findings support the effectiveness of EGIP as a unified interaction protocol for improving clinical-agent performance through controlled evidence access and acquisition.

cs.CL↗

Jev-LiteMAS: Separating Decisions from Generation in Multi-Agent Workflows

Multi-agent LLM systems rely on generative models for expert execution, workflow planning, and artifact review. Yet many coordination operations require only bounded decisions, such as selecting the next expert or judging artifact quality. Profiling three representative MAS frameworks, we find that routing and review account for 31.7-43.5% of generative-LLM tokens. However, replacing these operations with low-cost decision models introduces two interface gaps: generative planners produce execution plans and handoff instructions (planning gap), while reviewers provide actionable feedback for repairing flawed artifacts (review gap), which bounded decisions alone cannot provide. We introduce Jev-LiteMAS, a multi-agent framework that bridges both gaps by combining a lightweight decision model (Jev) with generative execution and feedback. To address the planning gap, Jev selects experts dynamically from the task and prior expert outputs, using accumulated context to ground subsequent execution without a pre-generated trajectory. To address the review gap, Jev screens intermediate artifacts against task criteria: passing artifacts proceed directly, whereas flagged ones undergo LLM re-review for acceptance or corrective feedback. Across four benchmarks and two backbones, Jev-LiteMAS reduces generative-LLM tokens by 12.7-54.6% over LAS, maintaining accuracy within 1.34 percentage points and lowering latency in seven of eight settings. Grounding routing in substantive expert outputs prevents up to 18.0 points in accuracy loss.

cs.AI↗

Proteus: A Self-Evolving Red Team for Agent Skill Ecosystems

Agent skills extend LLM agents with reusable instructions, tool interfaces, and executable code, and users increasingly install third-party skills from marketplaces, repositories, and community channels. Because a skill exposes both executable behavior and context-setting documentation, its deployment risk cannot be measured by single-shot audits or prompt-level red teams alone: a realistic attacker can use audit and runtime feedback to repeatedly rewrite the skill. We frame this risk as \emph{adaptive leakage} -- whether a budgeted attacker can iteratively revise a skill until it passes audit and produces verified runtime harm -- and present \ours{}, a grey-box self-evolving red-team framework for measuring it. Proteus searches a formalized five-axis skill-attack space. Each candidate is evaluated through a unified audit-sandbox-oracle pipeline that returns structured audit findings and runtime evidence to guide cross-round mutation. Beyond initial evasion, Proteus performs path expansion, which finds alternative implementations of successful attacks, and surface expansion, which transfers learned implementation patterns to new attack objectives beyond the original seed catalogue. Across eight phase-1 cells, Proteus reaches 40--90\% Attack Success Rate at $5$ rounds (ASR@5) with positive learning-curve slopes on both evaluated auditors. Phase-2 path/surface expansion produces 438 jointly bypassing and lethal variants, with SkillVetter bypassed at $\geq 93\%$ in every cell and AI-Infra-Guard, the strongest public auditor we evaluate, still admitting up to 41.3\% joint-success. These results show that current skill vetting substantially underestimates residual risk when evaluated against adaptive, feedback-driven attackers.

cs.CR↗

MAS-Shield: A Defense Framework for Secure and Efficient LLM MAS

Large Language Model (LLM)-based Multi-Agent Systems (MAS) are susceptible to linguistic attacks that can trigger cascading failures across the network. Existing defenses face a fundamental dilemma: lightweight single-auditor methods are prone to single points of failure, while robust committee-based approaches incur prohibitive computational costs in multi-turn interactions. To address this challenge, we propose \textbf{MAS-Shield}, a secure and efficient defense framework designed with a coarse-to-fine filtering pipeline. Rather than applying uniform scrutiny, MAS-Shield dynamically allocates defense resources through a three-stage protocol: (1) \textbf{Critical Agent Selection } strategically targets high-influence nodes to narrow the defense surface; (2) \textbf{Light Auditing} employs lightweight sentry models to rapidly filter the majority of benign cases; and (3) \textbf{Global Consensus Auditing} escalates only suspicious or ambiguous signals to a heavyweight committee for definitive arbitration. This hierarchical design effectively optimizes the security-efficiency trade-off. Experiments demonstrate that MAS-Shield achieves a 92.5\% recovery rate against diverse adversarial scenarios and reduces defense latency by over 70\% compared to existing methods.

cs.MA↗

Stop When Memory Suffices: Evidence-Conditioned Progressive Execution for LLM Agents

The continued development of LLMs toward persistent and adaptive intelligence increasingly requires long-term memory mechanisms that preserve and reuse information across interactions. Existing memory systems either compress and structure histories for efficient access or perform deep research over broader trajectories. The former lowers online cost but may omit temporal, causal, or cross-step dependencies, while the latter improves evidence coverage at substantial latency and inference cost. This raises a key question: can a memory system achieve strong answer quality while maintaining low online latency? We introduce Router-Mem, an evidence-conditioned progressive execution framework for long-horizon agent memory. Router-Mem first applies a shared low-cost retrieval prefix to obtain evidence. A lightweight sufficiency router then predicts whether the context supports early termination, which enable a single-token decision at inference time. It is trained with evidence-level supervision and rationale-conditioned representation distillation. When evidence is insufficient, Router-Mem reuses retrieval hits to expand memory blocks and perform deeper analysis and aggregation. Experiments on AMA-Bench and BEAM show that Router-Mem achieves 55.17\% and 38.77\% score while reducing average inference time by 27.3\% and 25.5\% compared with full memory execution.

cs.LG↗

BRA-Audit: Budgeted Runtime Auditing for LLM Multi-Agent Systems via Cumulative-Exposure Audit-Point Placement

LLM-based multi-agent systems (LLM-MAS) solve complex tasks through specialized collaboration, but inter-agent dependencies can propagate hallucinated or malicious outputs into system-level failures. Auditor agents mitigate these risks, yet existing strategies face an efficiency dilemma: end-only auditing reviews long trajectories and final outputs, potentially weakening audit effectiveness and enlarging rollback scope, while auditing every agent each round improves detection and localization at high token cost. How can guard performance be preserved while minimizing token cost? To address this problem, we propose BRA-Audit, a budget-aware runtime auditing framework that models MAS execution as a dynamic dependency graph and formulates audit scheduling as audit-point placement under a fixed audit-call budget to minimize cumulative unchecked exposure. Its greedy scheduler prioritizes influential and long-unaudited regions, while trusted audit points enable localized recovery. Across structured coordination, complex reasoning, and open-ended tasks, BRA-Audit restores performance close to the clean setting, remains competitive with heavy guard methods and reduces end-to-end token consumption by \(17.2\%\)--\(40.6\%\).

cs.MA↗

The Next Generation Virgo Cluster Survey (NGVS). II. A Catalog of Galaxies in the Virgo Cluster

The Next Generation Virgo Cluster Survey (NGVS) is a deep, high resolution imaging campaign that used the 1 deg$^2$ MegaCam instrument on the Canada-France-Hawaii Telescope to carry out a comprehensive optical survey of the Virgo cluster, from its core to its virial radius. The NGVS covers a contiguous area of 104 deg$^2$ (8.63 Mpc$^2$ at the 16.5 Mpc distance of Virgo) in the $u^*$-,$g$-,$i$-, and $z$-band, with additional limited coverage in $r$. In this paper, we present the final catalog of Virgo galaxies across the entire NGVS area. The catalog includes 3680 galaxies considered to be $bona~fide$ members of the cluster, spanning a factor of 2.5 million in luminosity, from $g = 8.42$ mag to $g = 24.41$ mag ($M_g = -22.67$ mag to $M_g = -6.68$ mag). With 2100 previously uncataloged galaxies, the NGVS catalog augments the number of known Virgo members by a factor 2.3. The catalog is complete down to $g = 18.6$ mag ($M_g=-12.5$ mag, corresponding to a stellar mass $M_* \sim 1.6\times10^7~M_{\odot}$ for an old stellar population) and 50% complete at $g = 22.0$ mag ($M_g=-9.1$ mag, $M_* \sim 6.2\times10^5~M_{\odot}$), three magnitudes deeper than the venerable Virgo Cluster Catalog (VCC), which for over 40 years has served as the reference standard for Virgo. Photometric and structural parameters are derived for all NGVS galaxies and presented in a series of tables, alongside nuclear and morphological classification, as well as stellar masses and, when available, radial velocities.

astro-ph.GA↗

E-mem: Multi-agent based Episodic Context Reconstruction for LLM Agent Memory

The evolution of Large Language Model (LLM) agents towards System~2 reasoning, characterized by deliberative, high-precision problem-solving, requires maintaining rigorous logical integrity over extended horizons. However, prevalent memory preprocessing paradigms suffer from destructive de-contextualization. By compressing complex sequential dependencies into pre-defined structures (e.g., embeddings or graphs), these methods sever the contextual integrity essential for deep reasoning. To address this, we propose E-mem, a framework shifting from Memory Preprocessing to Episodic Context Reconstruction. Inspired by biological engrams, E-mem employs a heterogeneous hierarchical architecture where multiple assistant agents maintain uncompressed memory contexts, while a central master agent orchestrates global planning. Unlike passive retrieval, our mechanism empowers assistants to locally reason within activated segments, extracting context-aware evidence before aggregation. Evaluations on the LoCoMo benchmark demonstrate that E-mem achieves over 54\% F1, surpassing the state-of-the-art GAM by 7.75\%, while reducing token cost by over 70\%.

cs.AI↗

OEP: Poisoning Self-Evolving LLM Agents via Locally Correct but Non-Transferable Experiences

Memory-augmented large language model (LLM) agents use iterative reflection and self-evolution to solve complex tasks, but these mechanisms introduce security risks. Existing agentic memory attacks require privileged access or explicit malicious content, making them detectable by advanced safety filters. This leaves a subtler attack surface underexplored: whether adversaries can induce agent to generate experiences that appear locally correct and semantically plausible yet induce harmful generalization during reflection. We find that reflective agents are vulnerable to such clean experiences, especially when paired with severe but plausible hypothetical consequences. Based on this observation, we introduce Obsessive Experience Poisoning (OEP), a low-privilege black-box attack requiring no direct control over the system prompt or memory database. OEP constructs adversarial clean edge-cases that combine locally correct solutions, non-transferable methods, and severe consequences, biasing reflection toward risk-averse rule formation. During memory consolidation, agents may over-trust self-generated reflections and distill localized experiences into high-priority but over-generalized rules, causing downstream failures. Evaluations across three domains show that OEP achieves ASR above 50\% with GPT-4o agents, and outperforms existing attacks under LLM auditing defense.

cs.CR↗

The Next Generation Virgo Cluster Survey (NGVS). XL. The Morphological Classification of Virgo Cluster Galaxies

We present a study of morphologies, based on deep $u^{*}g^{\prime}i^{\prime}z^{\prime}$ imaging of the Virgo Cluster from the Next Generation Virgo Cluster Survey (NGVS), for 3689 Virgo cluster members spanning a mass range of $\sim$$10^{11}M_{\odot}$ to $\sim$$10^5~M_{\odot}$. Our analysis introduces a new, two-component visual classification scheme developed to capture the morphological diversity of galaxies over more than six orders of magnitude in stellar mass. Our morphological classifications use two parameters to describe the global structure and star formation activity of each galaxy. Structural sub-codes denote features such as spiral arms, bars, disks, shells, streams, while star formation sub-codes indicate the form and location of the current star formation activity (e.g., in cores, clumps, filaments, etc). These visual classifications rely on deep $g^\prime$-band images, supplemented by $u^{*}g^{\prime}i^{\prime}$ color images, as well as unsharp-masked images for a subset of objects. We compare our classifications to previous results for bright member galaxies that used more established schemes, finding good agreement. We also measure quantitative classification statistics (e.g., CASGM$_{20}$) for a subset of the brighter galaxies, and present catalogs for some galaxy types of special interest, including structurally compact galaxies, ultra-diffuse galaxies, candidate ultra-compact dwarf transition objects, as well as candidate post-merger systems. These morphological classifications may be useful as a training set in the application of machine learning tools to the next generation of wide-field imaging surveys.

astro-ph.GA↗

Galaxy Model Subtraction with a Convolutional Denoising Autoencoder

Galaxy model subtraction removes the smooth light of nearby galaxies so that fainter sources (e.g., stars, star clusters, background galaxies) can be identified and measured. Traditional approaches (isophotal or parametric fitting) are semi-automated and can be challenging for large data sets. We build a convolutional denoising autoencoder (DAE) for galaxy model subtraction: images are compressed to a latent representation and reconstructed to yield the smooth galaxy, suppressing other objects. The DAE is trained on GALFIT-generated model galaxies injected into real sky backgrounds and tested on real images from the Next Generation Virgo Cluster Survey (NGVS). To quantify performance, we conduct an injection-recovery experiment on residual images by adding mock globular clusters (GCs) with known fluxes and positions. Our tests confirm a higher recovery rate of mock GCs near galaxy centers for complex morphologies, while matching ellipse fitting for smooth ellipticals. Overall, the DAE achieves subtraction equivalent to isophotal ellipse fitting for regular ellipticals and superior results for galaxies with high ellipticities or spiral features. Photometry of small-scale sources on DAE residuals is consistent with that on ellipse-subtracted residuals. Once trained, the DAE processes an image cutout in $\lesssim 0.1$ s, enabling fast, fully automatic analysis of large data sets. We make our code available for download and use.

astro-ph.IM↗

A Supermassive Black Hole in a Diminutive Ultra-compact Dwarf Galaxy Discovered with JWST/NIRSpec+IFU

The integral-field unit mode of the Near-Infrared Spectrograph (NIRSpec+IFU) mounted on the James Webb Space Telescope has now enabled kinematic studies of smaller and less massive compact stellar systems in which to search for central massive black holes (BHs) than ever before. We present here the first such detection using NIRSpec+IFU in its highest resolution (R~2700) mode. We report a $3σ$ detection of a central black hole with mass ${\cal M}_{BH}=2.2\pm1.1\times10^6\,M_\odot$ in UCD736 orbiting within the Virgo galaxy cluster based on Schwarzschild's modeling of the 1D kinematic profile. The presence of such a massive BH strongly argues against a globular cluster origin of this UCD, and rather suggests a tidally stripped formation route from a former $\gtrsim10^9\,M_\odot$ dwarf galaxy host. Two other methods produce results consistent with Schwarzschild's modelling, but can only provide upper-limits on ${\cal M}_{BH}$. This represents the detection of a BH in the most compact ($r_h\approx15\,{\rm pc}$) stellar system to date, with a ${\cal M}_{BH}$ corresponding to ~9 percent of the system's stellar mass, roughly in line with previously reported UCD BH detections and comparable to the BH detected in the compact elliptical galaxy NGC4486B.

astro-ph.GA↗

A JWST View of the Overmassive Black Hole in NGC 4486B

We present a new stellar dynamical measurement of the supermassive black hole (SMBH) in the compact elliptical galaxy NGC 4486B, based on integral field spectroscopy with JWST/NIRSpec. The two-dimensional kinematic maps reveal a resolved double nucleus and a velocity dispersion peak offset from the photometric center. Utilizing two independent methods-Schwarzschild orbit-superposition and Jeans Anisotropic Modeling-we tightly constrain the black hole mass by fitting the full line-of-sight velocity distribution. Our axisymmetric Schwarzschild models yield a best-fit black hole mass of $M_{BH} = 3.6^{+0.7}_{-0.7} \times 10^8 \, M_{\odot}$, slightly lower but significantly more precise than previous estimates. However, since our models do not account for the non-equilibrium nature of the double nucleus, this value may represent a lower limit. Across all tested dynamical models, the inferred $M_{BH}/M_*$ ratio ranges from ~ 4-13%, providing robust evidence for an overmassive SMBH in NGC 4486B. Combined with the galaxy's location deep within the Virgo Cluster, our results support the interpretation that NGC 4486B is the tidally stripped remnant core of a formerly massive galaxy. As the JWST/NIRSpec field of view is insufficient to constrain the dark matter halo, we incorporate archival ground-based long-slit kinematics extending to 5 arcsec. While this provides some leverage on the dark matter content, the constraints remain relatively weak. We place only an upper limit on the dark matter fraction, with $M_{DM}/M_{*} < 0.5$ within 1 kpc-well beyond the effective radius. The inferred black hole mass remains unchanged with or without a dark matter halo.

astro-ph.GA↗

Unveiling the nature and fate of the almost-dark cloud AGC 226178 through HI mapping

The origin of extragalactic, almost dark HI clouds with extreme gas-to-stellar mass ratios remains poorly understood. We investigate the nature and fate of the "almost dark" cloud AGC 226178, projected within the Virgo cluster, with an HI-to-stellar mass ratio of ~1000. We present deep single-dish HI mapping from the Five-hundred-meter Aperture Spherical Telescope (FAST), complemented by high-resolution interferometric data from the Very Large Array (VLA), as part of the Atomic gas in Virgo Interacting Dwarf galaxies (AVID) project. These observations provide the highest-quality HI analysis to date of such a cloud, combining resolution and sensitivity. FAST data reveal a short, low-velocity tail toward the dwarf galaxy VCC 2034, previously proposed as a possible origin for AGC 226178. However, VCC 2034 shows a line-of-sight asymmetric HI feature and cometary morphology indicating a stripping event unrelated to AGC 226178. VLA data reveal a velocity gradient across AGC 226178 and a clumpy internal structure. The velocity dispersion exceeds the thermal linewidth, implying turbulence or unresolved motions. The cloud cannot be gravitationally bound by atomic gas alone. The resolved HI clumps follow standard HI mass-star formation rate and mass-size relations, with those forming stars reaching surface densities above the threshold for self-shielding. We conclude that AGC 226178 is a free-floating HI cloud of unknown origin. The system appears to be in the process of disintegration. It is likely located well outside the Virgo cluster, as the preservation of its extended HI morphology within the cluster environment would otherwise require a substantial reservoir of unseen molecular gas with a mass exceeding that of the observed HI content. While confinement pressure from the hot intracluster medium may aid its stability, it is unlikely to be the dominant factor preventing its disruption.

astro-ph.GA↗