Search arXiv⌕ Search

arXiv subjects

Yang Zhang

Publications and source records attributed to Yang Zhang.

At least 37 records · Page 2Linked to original sources
arXiv2026

Learning to Steer LLM Agents and Multi-Agent Systems

Large language model (LLM)-based agents and multi-agent systems increasingly tackle complex tasks through long-horizon tool use and inter-agent collaboration. However, these systems are prone to deviating from the overall goal as execution unfolds, which wastes execution resources and leads to undesired outcomes, a failure termed goal drift. In this work, we propose a steering agent that supervises execution on the fly and steers the system back on track when drift occurs. The steering agent observes the execution context, deciding when to intervene with a low-cost fast path and how to intervene with a high-cost slow path. To unlock its steering intelligence, the ability to assess goal drift and generate effective interventions, we jointly train the two paths with reinforcement learning, combining a pairwise ranking objective with group-relative policy optimization (GRPO). Experiments on information-seeking search tasks show that the steering agent provides effective supervision for both single-agent and multi-agent systems, improving their performance and efficiency. The learned steering intelligence generalizes to out-of-domain tasks, transfers to unseen executors, and supports controllable intervention frequency without retraining.

cs.LG↗
arXiv2026

Grain Boundary Anisotropy and Its Influence on Helium Bubble Nucleation, Growth, and Decohesion in Polycrystalline Iron

The accumulation of helium bubbles at grain boundaries (GBs) critically degrades the mechanical integrity of structural materials in nuclear reactors. While GBs act as sinks for radiation-induced defects, their inherent structural anisotropy leads to complex helium bubble evolution behaviors that remain poorly understood. This work integrates accelerated molecular dynamics simulations and a novel atomic-scale metric, the flexibility volume (V_f), to establish the interplay between GB character, helium segregation, and bubble growth mechanisms in body-centered cubic iron. We demonstrate that V_f, which incorporates both local atomic volume and vibrational properties, qualitatively predicts deformation propensity. Our results reveal that the atomic-scale segregation energy landscape dictates initial helium clustering and subsequent bubble morphology, with low-energy channels in tilt Σ5 boundary facilitating one-dimensional migration while isolated deep traps in twist Σ13 boundary promote larger, rounder bubble morphology. Critically, besides gradual bubble growth via trap mutation mechanism, we identify two distinct stress-relief mechanisms: loop punching in anisotropic tilt Σ5 boundary and interfacial decohesion in twist Σ11 boundary, with the dominant pathway determined by the interplay between bubble morphology and local mechanical softness. This study establishes a fundamental connection between GB crystallographic and energetical anisotropy and helium bubble evolution, providing critical insights for designing radiation-tolerant microstructures.

cond-mat.mtrl-sci↗
arXiv2026

WinARDet: Compressed Window Auto-Regression for Long-Horizon Vision-Language Grounding

MLLM-based detectors formulate vision-language grounding as autoregressive coordinate generation, yet in dense scenes, they often collapse into a box repetition loop, repeatedly emitting identical boxes until the decoding budget is exhausted. We trace this failure to historical-answer attention takeover: Under full-history autoregression, repeated coordinate tokens in the growing answer history attract disproportionate attention, crowding out the attention sinks needed to attend to new visual evidence. % To break this loop, in this paper, we propose \textbf{Compressed Window Auto-Regression} (WinAR). This new mechanism partitions a long answer sequence into consecutive generation windows and restricts full token-level autoregression to the current window only while compressing each previous window into a fixed set of latent tokens that condition subsequent decoding. This design reduces the repeated exposure to raw historical coordinates that triggers the attention takeover, yet preserves cross-window dependencies essential for coherent multi-object generation. On RSDense, which contains 200-2,000 objects per image, our model WinARDet improves RexOmni F1-score (IoU=0.5) from 32.16 to 39.36 over the baseline, while reducing the redundancy rate from 35.48% to 13.00%. WinARDet also generalizes across decoding strategies and output formats, including parallel box decoding and polygon generation, providing a practical paradigm for long-horizon vision-language grounding across scene densities.

cs.CV↗
arXiv2026

On the Fomin-Kirillov algebras

The Fomin-Kirillov algebras $\mathcal{FK}_n$ are shown to be infinite-dimensional for $n\geq 6$. By computer calculations, we find an explicit nonzero primitive element of degree 14 in the identity component of $\mathcal{FK}_6$. For every $n\geq 6$, the existence of such an element implies that $\mathcal{FK}_n$ contains a polynomial subalgebra and is therefore infinite-dimensional, and that the canonical homomorphism from $\mathcal{FK}_n$ onto its associated Nichols algebra is not an isomorphism.

math.RA↗
arXiv2026

GeoCloak: An Agentic Framework for Semantics-Preserving Geolocation Privacy via Peripheral Clue Manipulation

The rapid advancement of large vision language models (LVLMs) has made geolocation inference from images highly accurate and accessible. This poses serious privacy risks: a passive adversary with black-box LVLM access can infer a user's location from a single photo with no metadata. Existing protection methods rely almost exclusively on adversarial perturbation, which introduces perceptible visual artifacts and utility degradation. We identify the visual-geolocation asymmetry: geolocation-decisive clues, such as street signs, tend to be visually peripheral but dominate location inference. Exploiting this asymmetry, we propose GeoCloak, an agentic framework for semantics-preserving geolocation privacy protection. It acts as a clue-aware planner that analyzes each image to identify geolocation-relevant clues, dispatches targeted semantic edits to directly suppress peripheral clues, and autonomously configures adversarial perturbation to further suppress residual geolocation features, all under iterative utility-aware feedback. We evaluate GeoCloak on four diverse datasets spanning 108 countries and 576 cities against five LVLMs and three inference methods. GeoCloak reduces overall city-level accuracy from 41.2% to 9.9%, approximately half of the best-performing baseline, while achieving substantially superior utility preservation both qualitatively and quantitatively. We hope this work takes a meaningful step toward practical, semantics-preserving geolocation privacy protection, giving users greater control over the location information inadvertently embedded in their shared images.

cs.CV↗
arXiv2026

Graph-Reweighted Behavior Cloning with Human Interventions for Precise Robotic Manipulation

Long-horizon precise manipulation produces mixed-quality data for post-training vision-language-action (VLA) models: even successful demonstrations and intervention-assisted rollouts may contain detours, hesitations, failed attempts, and imperfect recoveries. The key challenge is to identify which action chunks actually advance the task, rather than uniformly cloning or discarding imperfect trajectories. We propose Graph-Reweighted Behavior Cloning (GR-BC), an offline iterative post-training framework that builds a task-conditioned goal-reachability graph over demonstrations and iteratively collected rollouts, including both successful and failed episodes with human interventions. Reverse graph search computes minimal remaining steps to task success within the dataset support, and each action chunk is weighted by how much its execution reduces this potential. This consequence-based weighting reuses recoverable segments from failed trajectories, leverages interventions as recovery paths, and down-weights locally regressive chunks even inside successful trajectories. New rollouts are incorporated by appending them to the graph and rerunning search, without training a reward model, action-conditioned critic, or progress predictor. Across four precise manipulation tasks and three VLA backbones, GR-BC improves single-round supervised fine-tuning and reaches 91% success after a few offline iterations, without trajectory comparisons or progress labels. More details and videos are available at: https://gr-bc.github.io

cs.RO↗
arXiv2026

Photoacoustic imaging of human brain function through intact adult skulls

Blood-oxygen-level-dependent (BOLD) functional magnetic resonance imaging (fMRI) has revolutionized noninvasive imaging of human brain function, but it indirectly senses deoxyhemoglobin through water-proton relaxation and relies on bulky, costly, noisy, enclosed, superconducting-magnet scanners. In contrast, photoacoustic computed tomography (PACT) directly probes hemoglobin absorption using a more compact, lower-cost, quieter, open, magnet-free system. However, functional PACT of the adult human brain has so far been demonstrated only after surgical skull removal, because the skull strongly attenuates optical and acoustic signals. Here, we demonstrate non-invasive transcranial functional PACT (T-fPACT) through the intact adult skull. A densely packed, high-channel-count (3,072-element) hemispherical ultrasonic array provides near-$2π$-steradian simultaneous acquisition, enabling high sensitivity and single-shot three-dimensional imaging at 20 Hz that eliminates motion blurring. Participant-specific stabilization further enables measurement of small task-evoked hemodynamic responses over the entire task duration. In healthy adults, T-fPACT localized responses to expected sensorimotor regions, distinguished task-specific motor representations, and yielded consistent localizations across repeated task runs and visits. Comparison with high-field (7 T) fMRI showed spatially corresponding functional regions, while dual-wavelength imaging revealed distinct hemoglobin-weighted responses. These findings overcome the long-standing skull barrier in human brain PACT and establish non-invasive, rapid volumetric photoacoustic imaging of brain function.

physics.med-ph↗
arXiv2026

Write Once, Plan Many: Building Reusable Visual Planning Systems from One Scene

Vision-language models (VLMs) can reason about visual planning tasks, but querying them for every instance incurs repeated inference cost and latency. Generating reusable planning programs offers an alternative, but constructing reliable planning systems for new problem domains without extensive supervision remains challenging. We present a training-free framework that constructs reusable perception and planning programs from a domain specification and a single practice scene without a paired solution. The central challenge is obtaining useful feedback from the limited input. Our framework addresses it by leveraging VLM's ability to provide a renderer and a dynamic model based on its scene understanding. Rendering detected states enables comparison with the practice scene, while checking simulated action outcomes against the domain specification helps expose errors in rule implementation. Together, these checks guide refinement of the grounding program and planner without requiring solution labels. After refinement and component selection, the programs are frozen and solve new instances without further VLM calls. The same framework constructs systems across domains without manual pipeline redesign. Across fifteen domain variants spanning 2D and 3D scenes with discrete and continuous state, our systems achieve 99.3% average planning success, compared with 80.9% for the strongest domain-level program-generation baseline and 90.7% for the strongest instance-level baseline. Cost analyses show a favorable success-cost trade-off, and ablations confirm the contributions of the executable domain model, execution feedback, and component selection.

cs.AI↗
arXiv2026

MemeTrace: A Multimodal LLM-Based Agent for Safety-Aware Meme Provenance Analysis

Online memes are often carriers of unsafe content such as disinformation and propaganda. Their rapid evolution in response to news events and strong dependence on contextual information make it difficult for platforms and moderators to detect and moderate them effectively. Moreover, existing moderation studies typically classify memes as safe or unsafe based on individual images, lacking the ability to perform broader provenance analysis, i.e., where a meme originates, how it spreads, what cultural symbolism it represents, whether it poses potential risks to society, etc. To bridge this gap, we introduce MemeTrace, a multimodal LLM-based agent that automatically traces the provenance and potential risks of emerging memes. We evaluate MemeTrace on Meme100, a dataset of 100 real-world memes, using both human annotation and automatic metrics. The results demonstrate that MemeTrace substantially outperforms the VLM baseline. More importantly, MemeTrace achieves higher accuracy in detecting unsafe memes, especially those involving disinformation and sensitive content, and remains robust under visual perturbations.

cs.AI↗
arXiv2026

Probing LZ-Motivated TeV Higgsino Dark Matter through Heavier Sparticles at LHC

The nearly pure TeV-scale Higgsino dark matter provides a natural explanation for the high-energy nuclear-recoil excess reported by LUX-ZEPLIN. Such a heavy and compressed Higgsino spectrum makes direct searches at the LHC challenging. We therefore investigate complementary probes through the production and decay of heavier sparticles. We perform a comprehensive analysis and obtain the following observations: (i) For a TeV Higgsino, the current LHC Run-2 searches exclude gluino up to approximately 2.2~TeV, while the HL-LHC can give an exclusion reach of 2.5--2.6~TeV and a discovery reach of 2.3--2.4~TeV; (ii) The light-flavor squark coannihilation window lies beyond the Run-2 reach, but its lower end could be excluded at the HL-LHC under optimistic assumptions; (iii) The Run-2 search does not exclude the stop or sbottom benchmarks; (iv) The slepton and heavier-electroweakino productions are too small to access at the HL-LHC.

hep-ph↗
arXiv2026

V-JEPA Policy: Building Effective World-Action Models on Predictive Visual Latents

World-action models (WAMs) couple future visual-state prediction with action generation. By adapting video generators or image-editing models pretrained at scale, a prominent line of recent WAMs inherits both predictive knowledge and the models in which it was learned. We ask whether a predictive visual latent space induced by large-scale predictive pretraining can instead provide a sufficient foundation for effective WAM learning without inheriting a complete pretrained visual generative model. To answer this question, we introduce V-JEPA Policy, a simple framework that builds a WAM on the latent space of a frozen V-JEPA 2.1 encoder. An instruction-conditioned future-latent predictor and a flow-matching action expert are jointly learned from scratch in a single downstream stage, with the predictor's future-informed context key--value states conditioning action generation. With 0.9B total parameters, of which 0.6B are trainable, V-JEPA Policy achieves competitive performance with representative WAM and vision-language-action baselines across LIBERO, LIBERO-Plus, and RoboCasa-GR1. Comparing visual foundations under the same downstream framework and training budget identifies V-JEPA latents as more effective than the discriminative, reconstructive, and video-understanding-oriented alternatives, particularly under distribution shifts. Beyond task-specific learning, pretraining the predictor on DROID video--instruction pairs without action labels and adapting it into a WAM yields substantial gains in downstream control and out-of-distribution generalization. Together, these findings establish predictive visual latents as a foundation for effective WAM learning from task-specific demonstrations and for transferring future-modeling knowledge acquired from broader in-the-wild videos. Project page: https://breez3young.github.io/vjepa-policy/

cs.CV↗
arXiv2026

Quantum Geometry and Pair Breaking in Intra- and Intervalley Superconductivity of Rhombohedral Graphene

The observation of superconductivity emerging from a spin- and valley-polarized normal state in rhombohedral graphene raises a basic puzzle: intravalley Cooper pairs involve states that are not generically degenerate because of trigonal warping, and therefore lack the particle--particle nesting underlying the conventional Cooper instability. While recent model studies suggest that nontrivial band geometry can strongly promote chiral pairing, how this geometric enhancement competes with realistic band-structure-induced pair breaking remains unclear. Here we study Kohn--Luttinger superconductivity in electron-doped rhombohedral tetralayer graphene using a DFT-fitted multiorbital tight-binding model, retaining the full Bloch-wave-function form factors in both RPA screening and the projected Cooper interaction. We separate the superconducting instability into three contributions: the symmetry-resolved form-factor-dressed interaction, the Cooper-pair phase space determined by the symmetrized pair dispersion, and the energy mismatch generated by trigonal warping. We find that the Bloch form factors strongly favor intravalley pairing, with a leading predominantly $p_x-ip_y$ instability and a substantially larger maximum $T_c$ than the intervalley reference state. At the same time, trigonal-warping-induced mismatch cuts off the Cooper logarithm and confines intravalley superconductivity to a much narrower region of the phase diagram. The dominant instability occurs near pair-Lifshitz transitions and primarily in reconstructed annular or three-lobed Fermi-surface regimes. Our results show that superconductivity in rhombohedral graphene is controlled by the combined effects of quantum-geometry-dressed interactions, pair fermiology, and trigonal-warping-induced pair breaking.

cond-mat.supr-con↗
arXiv2026

Swarm Telepathy: Prefill-Free KV-Cache Transfer Across Families

LLM multi-agent systems split work across agents that communicate through text. An agent summarizes its findings, another agent receives the text. This constrains communication to discrete tokens and costs redundant compute, limiting bandwidth and capabilities of multi-agent systems. Recent works demonstrate KV-cache transfer between LLMs but they either require the LLMs to be within the same family or cannot function across context, thus requiring the receiver to prefill text. We present SwaT (Swarm Telepathy), the first cross-family KV-cache transfer method that allows prefill-free communication in multi-family multi-agent systems. In an agent-to-agent communication setting, combining SwaT with recent advances in KV-cache compaction beats text-to-text communication in both accuracy and speed, with a mean of $40\%$ higher accuracy and a mean of $4\times$ faster wall-clock time. Experiments show SwaT achieves high performance on 17 benchmarks across 12 LLMs and 6 families. Additionally, given a sender Vision-Language Model (VLM), SwaT is able to transfer the vision capabilities to a text-only receiver LLM exceeding text-to-text communication despite SwaT being trained on text-only data. SwaT keeps both sender and receiver models frozen without fine-tuning and only requires $<50M$ tokens and $<1$ GPU-day to train.

cs.LG↗
arXiv2026

Contrasting $Γ$- and K-Valley Moiré Physics in Twisted Monolayer/Bilayer WSe$_2$

Electronic orbital character plays a central role in determining electronic correlations, spin-orbit coupling, dimensionality, and ultimately the quantum phases of condensed-matter systems. Two-dimensional moiré materials have emerged as highly tunable platforms for exploring correlated phenomena, but the role of orbital degrees of freedom remains largely unexplored. Here, we identify twisted monolayer/bilayer WSe$_2$ as a platform in which displacement-field tuning enables moiré physics to be realized in both the $K$ and $Γ$ valleys. The distinct orbital characters of these valleys give rise to contrasting correlated phases at moiré filling factors $ν=1$ and $ν=1/3$. At $ν=1$, the $K$-valley state is a weak insulator, consistent with an antiferromagnetic state near a van Hove singularity in the intermediate-coupling regime, similar to that observed in twisted bilayer WSe$_2$. In contrast, the $Γ$-valley state exhibits a pronounced Pomeranchuk effect, consistent with proximity to a Mott transition. At $ν=1/3$, the $K$ valley hosts a robust generalized Wigner crystal, whereas the $Γ$-valley state lies near the crystallization boundary and again exhibits a Pomeranchuk effect, with localization enhanced by increasing temperature or magnetic field. Our work highlights the importance of orbital character in defining quantum phases in moiré systems, and identify the $Γ$ valley as a promising platform for exploring correlated phenomena near quantum phase transitions, where competing phases and enhanced fluctuations may give rise to unconventional phases.

cond-mat.mes-hall↗
arXiv2026

Overlapping Subcritical Bubbles: Free Energy and Lifetime

Subcritical bubbles can form an appreciable population during weak first-order phase transitions, but are usually treated as isolated fluctuations. This raises the question of whether spatial overlap between neighboring subcritical bubbles can modify their evolution. We address this question by combining analytic free-energy calculations for composite Gaussian profiles with Langevin simulations of overlapping configurations. We find that the overlap lowers the free-energy cost and generally increases the lifetime of subcritical bubbles, with the enhanced persistence potentially feeding back on their abundance. Such overlap can therefore generate collective effects and should be incorporated into kinetic descriptions of subcritical-bubble populations.

hep-ph↗
arXiv2026

Twist-angle-dependent quantum phase diagrams in twisted bilayer MoTe2

Two-dimensional semiconducting moiré superlattices provide a highly tunable platform for investigating strongly correlated and topological quantum phases. As a prototypical example, twisted bilayer MoTe2 (tMoTe2) has been shown to host fractional topological phases at zero magnetic field. However, their evolution with twist angle and competition with other correlated quantum states remain largely unexplored. Here we report a systematic transport study of twist-angle-dependent phase diagrams in tMoTe2 across a range of 3.79°-5.76°, revealing an evolution from valley-polarized, fractional topological states to valley-degenerate correlated states and superconductivity. At relatively small twist angles, partially-filled Chern bands of tMoTe2 host fractional quantum anomalous Hall (FQAH) states following the Jain sequence, together with signatures of an anomalous composite Fermi liquid at moiré hole filling factor νh = 1/2. Increasing twist angle progressively suppresses fractional topological phases and reconstructs the half-filled Chern band into symmetry-breaking integer Chern insulating states. At νh = 1, we observe a transition from robust integer quantum anomalous Hall (IQAH) insulators at small angles to displacement-field-tuned, topologically trivial correlated insulators at larger angles. Remarkably, at twist angles above approximately 5.6°, superconductivity emerges adjacent to the νh = 1 correlated insulator, with a phase diagram closely resembling that recently reported in twisted bilayer WSe2. Our results uncover a unified twist-angle-driven phase evolution linking fractional topology, symmetry breaking, magnetic order, and superconductivity, providing new insight into the emergent quantum phenomena in moiré systems.

cond-mat.mes-hall↗
arXiv2026

TopoGraphRAG-Bench: Evaluating Multimodal GraphRAG on Layout-Grounded Evidence Reasoning

Real-world documents distribute evidence across text, tables, figures, and captions within complex page layouts. Answering complex questions over such documents therefore requires more than retrieving relevant passages: systems must recover the evidence topology that connects heterogeneous evidence units. Existing GraphRAG evaluations remain largely text-centered, while multimodal document RAG benchmarks assess cross-modal retrieval and generation without directly evaluating recovery of the intended evidence topology. We introduce TOPOGRAPHRAG-BENCH, a layout-grounded benchmark for multimodal evidence reasoning in GraphRAG, comprising 2,024 questions over 201 long, visually rich documents. Questions are constructed bottom-up from text, figure, and table evidence units under three controlled topologies: single-hop retrieval, bridge-chain reasoning, and multi-source synthesis. To ensure that questions preserve their intended structure, we apply counterfactual validation for shortcut resistance, modality necessity, and evidence necessity. We evaluate text-only GraphRAG, page-level visual retrieval, and multimodal GraphRAG systems using retrieval, generation, and topology-aware reasoning metrics. Multimodal GraphRAG systems achieve the strongest overall performance, but still fail when visual-textual evidence alignment or multi-unit composition is incomplete. Text-only GraphRAG struggles when key dependencies are grounded in figures or tables, while page-level visual retrieval lacks the fine-grained structure needed for topology recovery. These findings motivate GraphRAG systems that move beyond text-derived entity relation graphs to explicitly model document layouts, cross-modal evidence alignment, and the reasoning roles of evidence units. Code and data are available at https://richardlrc.github.io/TopoGraphRAG-Bench/.

cs.AI↗
arXiv2026

Stability and Diversity of Networked Self-Consuming Generative Ecosystems

The widespread deployment of generative AI has made it increasingly difficult to distinguish synthetic content from real data. Consequently, synthetic data is inevitably incorporated into the training pipelines of future model generations, forming a self-consuming training loop. Prior work has studied the effects of such recursive self-consuming training, but analyses have largely been limited to isolated models, where a model consumes only its own synthetic data, or to simplified interactions between two models. This paper takes a first step toward understanding networked self-consuming generative models, in which multiple models consume synthetic data generated by one another through complex interaction pathways. We introduce a theoretical framework representing models as nodes in a directed, weighted graph, with edge weights governing the flow of synthetic data among models. Using this framework, we analyze the long-term behavior of networked models under retraining dynamics, establishing conditions for convergence and characterizing the resulting fixed points. We further investigate how the system's long-term stability and diversity are shaped by each model's access to real data, cross-model data consumption, and the structure of the interaction graph.

cs.LG↗