arXiv:2607.14287v1 Announce Type: new
Abstract: Defect segmentation in additive manufacturing (AM) X-ray computed tomography (XCT) images remains challenging due to severe class imbalance and large distribution shifts across scan conditions. Although recent foundation models such as the Segment Anything Model (SAM) provide strong general-purpose segmentation priors, their natural-image pre-training transfers poorly to the AM XCT domain, where defects appear as subtle non-semantic microstructural anomalies. Moreover, adapting SAM to the AM domain is further limited by the large domain gap and scarcity of labeled real XCT data. We present XCT-SAM, a sequential parameter-efficient adaptation framework for AM XCT defect segmentation. Instead of adapting SAM directly from natural images to XCT data, we first fine-tune Conv-LoRA adapters on an alloy-microstructure dataset and subsequently transfer the adapted model to XCT images, progressively bridging the domain gap. Using Conv-LoRA adapters with rank r=2, the framework injects convolutional spatial inductive bias into SAM's backbone while training approximately 4.15M parameters and keeping over 99% of the model frozen. We evaluate XCT-SAM on out-of-distribution CycleGAN-XCT benchmarks and real-world NIST XCT scans. Across both settings, XCT-SAM consistently outperforms zero-shot SAM and other domain-adapted SAM baselines, achieving the best overall IoU and Dice scores. These results demonstrate the effectiveness of intermediate domain adaptation with parameter-efficient adapters for industrial XCT defect segmentation. The source code is publicly available at https://github.com/Mahedi-61/XCT-SAM.git
Science Journals
arXiv:2511.04402v4 Announce Type: replace-cross
Abstract: Motivated by the ubiquity of decoherence in quantum hardware and the growing role of imaginary-time evolution (ITE) in quantum algorithms, we investigate how many-body correlations generated by imaginary-time filtering are modified by local decoherence. We introduce measurement-dressed imaginary-time evolution (MDITE), which alternates ITE with projective-measurement channels, producing a competition between low-energy filtering and local dephasing. By developing a new efficient quantum Monte Carlo method, we uncover MDITE mixed-state transitions with spontaneous-symmetry-breaking signatures in the driving of 1D transverse-field Ising and 2D columnar dimerized Heisenberg Hamiltonians in the resulting density matrices. In the continuous limit, the Choi-Jamiolkowski mapping yields a tractable equilibrium description with conformal criticality that qualitatively captures the phase transitions. At finite protocol parameters, however, the four-point correlator violates the conformal cross-ratio form and the critical exponents deviate from their continuous-limit values, signaling the loss of conformal symmetry and richer nonequilibrium criticality. Our results establish MDITE as a controlled setting for exploring mixed-state phases and critical phenomena driven by the interplay between imaginary-time filtering and decoherence.
arXiv:2607.14196v1 Announce Type: new
Abstract: This study presents OCELOT (Observation-Centric Estimation and Learning for Outlook Trajectories), a global machine-learning forecasting system that predicts future Earth observations directly from heterogeneous satellite and in-situ measurements. Unlike data-driven weather models trained on gridded reanalysis states, OCELOT operates natively in observation space, preserving instrument-specific sampling, viewing geometry, and measurement characteristics. The system combines per-instrument graph-attention encoders, a shared spherical icosahedral latent mesh, a hybrid sliding-window Transformer/spatial graph neural network processor, and metadata-conditioned decoders to produce forecasts up to 12 h ahead. OCELOT is trained on observations for the years 2015 through 2023, validated on the year 2024, and evaluated out of sample on 2025 observations across satellite radiances, radiosondes, aircraft, and surface networks. In the 2025 evaluation, OCELOT produces spatially coherent +12 h forecasts across independent observing systems: microwave temperature-sounding channels show RMSE values of 1.24-1.87 K, while the more surface- and cloud-sensitive AVHRR infrared window channel shows a higher RMSE of 3.95 K. Vertical profile diagnostics show physically consistent radiosonde and aircraft temperature structure. Surface forecasts remain stable through 12 h, with 2-m air-temperature RMSE increasing from about 3.2 K at +3 h to about 3.6 K at +12 h. In paired observation-space comparisons, OCELOT remains less accurate than operational GFS but substantially outperforms persistence at longer lead times for 2-m temperature and 10-m wind components. These results demonstrate that observation-space forecasting can recover large-scale atmospheric structure and provide meaningful short-range skill without reanalysis supervision.
arXiv:2607.12248v2 Announce Type: replace-cross
Abstract: Large pretrained time-series models such as TimesFM are attractive for financial forecasting, but raw directional accuracy is a misleading scoreboard in equity markets. An early LoRA adapter in this project appeared to reach roughly 80% directional accuracy; we show this is not evidence of skill. Over a long horizon in a rising market, a trivial "always-up" rule attains comparably high accuracy without using the input at all. To separate genuine skill from this base-rate artifact, we build a reproducible, frozen-data benchmark with expanding walk-forward folds, a stratified held-out-ticker split, honest baselines (zero-shot TimesFM, always-up, random-walk, persistence, AR(1)), and paired significance tests (McNemar, Diebold-Mariano) under Benjamini-Hochberg FDR control. We apply the identical method to two universes -- a tech-heavy NASDAQ-100 and a broad S&P 500 -- reporting excess accuracy over the always-up base rate. Three findings replicate. First, when the historical ~80% condition is recreated, the high number is a base rate of ~0.70 that the fine-tuned model scores below. Second, pooled LoRA shows no directional skill over the base rate at any horizon on either universe (negative at the six-month horizon). Third, per-sector specialization is significantly worse than a single pooled adapter (Diebold-Mariano p<0.001 on held-out stocks at h=128). Fine-tuning's only measurable benefit is a statistically significant reduction in point-forecast error relative to zero-shot TimesFM, which nonetheless does not beat naive baselines and confers no tradeable directional edge. The contribution is methodological: a defensible, fully seeded protocol that prevents the base-rate trap, together with the replicated negative result it produces.
arXiv:2607.12420v2 Announce Type: replace-cross
Abstract: The quantum Hall effect in gapped systems represents a defining signature of nontrivial topology. Realizing this principle in gapless matter has remained a central challenge in quantum materials. Chiral topological semimetals provide a unique platform to achieve this aim via symmetry-protected multifold crossings that act as Berry-curvature monopoles. When optical transitions are confined to a single multifold node, the resulting circular photogalvanic effect is predicted to be quantized in terms of the topological charge of the node. In real materials, however, this phenomenon remains experimentally elusive, obscured by trivial band transitions, the energy separation between the node pairs, and their relative positions with respect to the Fermi level. Here we observe a quantized circular photogalvanic effect in the chiral topological semimetal Rh0.95Ni0.05Si. Ni substitution opens a photon-energy window dominated by interband optical transitions at the {\Gamma}-point multifold node. This allows circularly polarized near- to mid-infrared pulses to drive a helicity-odd terahertz response that manifests three hallmarks of quantization: a sharp onset, a wavelength-independent plateau governed by the magnitude of monopole charge, and an abrupt cutoff imposed by Pauli blocking. Our work establishes an all-optical analogue for the quantum Hall effect and a new paradigm for topological quantization in gapless matter.
arXiv:2607.15227v1 Announce Type: new
Abstract: How different populations visually explore artworks bears on cognitive science and on accessibility design, yet most eye-tracking work in autism has used social scenes rather than art, and has analysed where the eyes land while ignoring when and in what order. We present a comparative free-viewing study across three groups, autistic adults (ASD), trained artists, and neurotypical observers, who each viewed 30 paintings for 15s. We introduce a directed, metric-grounded framework that compares groups along two complementary axes: a spatial axis, in which one group's fixation-density map predicts another's fixations under six saliency metrics (AUC-Judd, NSS, CC, SIM, KL, Information Gain); and a temporal axis, in which individual scanpaths are compared with MultiMatch, ScanMatch, a foveal-disc IoU score (FDISS), and dynamic time warping (DTW). Fixations are extracted uniformly for all groups with a dispersion-threshold algorithm. Three results converge. (i)Artists and neurotypicals are almost indistinguishable in both space (density-map correlation CC=0.96) and time (they form the most alignable scanpath pair), whereas ASD gaze diverges from both. (ii)ASD attention is dissociated: it matches artists' wide spatial exploration (dispersion, explored area) but carries a distinct temporal signature, shorter fixations, less dwell, and the most idiosyncratic (least self-consistent) scanpaths of any group. (iii)ASD gaze is not selectively artist-like on any metric; if anything it is marginally closer to neurotypical. Together these findings indicate that autistic viewing of art is a distinct, group-specific attentional profile in both space and time, and they motivate population-conditioned models of aesthetic attention. We release all analysis code and per-stimulus results.
arXiv:2607.05268v3 Announce Type: replace
Abstract: Hyperbolic vision-language models are designed to encode abstraction geometrically: general concepts near the origin, specific ones farther out, and entailment cones representing directed order. We ask whether trained MERU, HyCoCLIP, and PHyCLIP models actually use these mechanisms. We audit seven released checkpoints and matched from-scratch interventions, using diagnostics that distinguish active hyperbolic geometry from angular structure and supervision effects. All audited converged checkpoints remain near-Euclidean in the dimensionless radius $u=\sqrt{c}\rho$, which measures how strongly embeddings experience hyperbolic geometry: the largest observed image-side value is $0.37$ -- well below $u\approx0.84$, where local metric distortion reaches $10\%$. Releasing the curvature floor changes curvature and norms but not this regime, with mixed, generally modest downstream shifts. Trained entailment cones are saturated or nearly saturated, so low violation rates can arise from trivially wide cones rather than learned order. Preregistered semantic traversal detects weak within-branch order but no operative full-hierarchy readout. Shuffle-controlled tests detect no pair-specific radial ordering in released checkpoints, and no positive result is consistent across all three matched ViT-B seeds. We trace this to a low-curvature shortcut: lowering curvature widens entailment cones and suppresses violations without learning order. In the probed trajectories, gradient decomposition identifies entailment as the dominant curvature-lowering pressure during collapse. Yet curvature contracts even when entailment is removed, so the shortcut is not the sole cause. Under our diagnostics, the audited formulations do not demonstrate an operative radial or cone-based hierarchy. We distill the audit into a five-number geometry report for evaluating future hierarchy claims.
arXiv:2511.14441v2 Announce Type: replace-cross
Abstract: To distinguish Markov equivalent graphs in causal discovery, it is necessary to restrict the structural causal model. Crucially, we need to be able to distinguish cause $X$ from effect $Y$ in bivariate models, that is, distinguish the two graphs $X \to Y$ and $Y \to X$. Location-scale noise models (LSNMs), in which the effect $Y$ is modeled based on the cause $X$ as $Y = f(X) + g(X)N$, form a flexible class of models that is general and identifiable in most cases. Estimating these models for arbitrary noise terms $N$, however, is challenging. Therefore, practical estimators are typically restricted to symmetric distributions, such as the normal distribution. As we showcase in this paper, when $N$ is a skewed random variable, which is likely in real-world domains, the reliability of these approaches decreases. To approach this limitation, we propose SkewD, a likelihood-based algorithm for bivariate causal discovery under LSNMs with skewed noise distributions. SkewD extends the usual normal-distribution framework to the skew-normal setting, enabling reliable inference under symmetric and skewed noise. For parameter estimation, we employ a combination of a heuristic search and an expectation conditional maximization algorithm. We evaluate SkewD on novel synthetically generated datasets with skewed noise as well as established benchmark datasets. Throughout our experiments, SkewD exhibits a strong performance and, in comparison to prior work, remains robust under high skewness.
arXiv:2607.13276v2 Announce Type: replace
Abstract: Aurora DSQL is a serverless SQL database designed for cloud-scale transaction processing with multi-region active-active capabilities. Built on a disaggregated architecture, DSQL separates compute, storage, and transaction coordination into independent, horizontally scalable services. Query processors run in Firecracker MicroVMs executing PostgreSQL-compatible SQL without local state. The system uses multiversion concurrency control with precision timestamps for coordination-free reads and optimistic concurrency control for writes, deferring coordination to commit time through distributed adjudicators and the Journal replication system. This minimizes cross-region latency by requiring coordination only during commits, not individual statements. DSQL enables elastic scaling from zero to millions of transactions per second while providing strong consistency, ACID transactions, and continuous availability during availability zone or region failures.
arXiv:2607.13220v2 Announce Type: replace
Abstract: Most AI-for-science systems focus on scaling a single reasoning process by using better models, larger context windows, long-horizon agentic execution, or digital co-scientists working with one principal user. However, challenging scientific problems are rarely solved by one reasoner alone. They are solved by teams whose members carry different priors, experimental background, tacit knowledge, and domain-trained intuitions. The open problem is therefore not only how to scale models, but how to develop "networked intelligence", scaling the connections between humans and AI systems so that a result or hypothesis produced in one context reaches another person, agent, instrument or robot that can act on it. We introduce Mycelium, an active shared workspace that automatically connects researchers and AI agents. As human users and agents work, the system captures important observations and hypotheses, tracks how they relate to the team's evolving knowledge model, and routes them to the person or agent whose next decision they can inform. We evaluate Mycelium through a real-world scientific discovery use case: a biological multi-omics campaign where shared context turned a local analytical finding into a cross-expert mechanistic constraint and ultimately into an experimental design. Finally, we describe networked intelligence as sparse conditional computation over distributed scientific contexts. This framework establishes when a scaled standalone agent is sufficient, and when isolated data and specialized expertise make a networked approach essential.
arXiv:2607.14233v1 Announce Type: new
Abstract: Physics-informed neural networks (PINNs) have had a broad research impact in modeling domains governed by partial differential equations (PDE). However, PINNs have been shown to perform poorly, sometimes even converging to trivial solutions, in challenging PDE domains, or when generalizing to unseen but related PDE domains. Previously proposed solutions detail hyperparameter tuning to reduce loss imbalance between data-driven and physics guided losses, curriculum learning based training strategies, or dynamic re-sampling of hard collocation points. These methods face certain pitfalls: hyperparameter tuning is expensive, designing a training curriculum is ambiguous in multi-parameter PDE settings, and dynamic resampling still fails in complex PDE settings. Complementary to this line of thinking, we believe the initial PINN network weights also play a crucial role in the emergence of catastrophic failures during training, yet the effect of PINN weight initialization has been surprisingly under-investigated. To this end, we propose a framework for Learned Initialization via Gated Layerwise Optimization (LIGO-PINN) to overcome PINN convergence failures. Through rigorous evaluation on 1D and 2D PDE domains, including a challenging 2D fluid dynamics setting, we demonstrate that our methodology outperforms state-of-the-art methods designed to alleviate PINN failures, achieving a 91.5% average performance improvement across six baselines and 81% over the strongest baseline. We also verify that LIGO-PINN generalizes to 3D unstructured domains. Finally, we analyze training dynamics across all three PDE domains to explain both LIGO-PINN's improvement and the convergence failure of traditional PINNs. Code: https://github.com/scailab/ligo-pinn
Keywords: Machine Learning, Physics-Informed Neural Networks, Deep Learning, PDE Modeling
arXiv:2512.05166v2 Announce Type: replace-cross
Abstract: In practical applications, MXenes are often subjected to a variety of loads, including compression. While their mechanical response under different loading conditions, such as tensile loading, has been extensively studied, their compressive instability remains largely unexplored. The compressive and post-buckling behavior of Ti2C and Ti2CO2 MXene nanosheets is studied using molecular dynamics (MD) simulations and a nonlocal formulation. The employed interatomic potential is first validated against experimental and density functional theory (DFT) data for structural and mechanical properties. The results indicate that classical continuum mechanics underestimates the buckling strains, whereas the nonlocal formulation adequately captures the observed response. A systematic examination of various defect types up to a defect fraction of 3% reveals that while isolated point defects primarily reduce the critical buckling stress, vacancy clusters significantly alter the buckling mode shapes. Lateral confinement pressure and oxygen surface termination substantially increase the buckling stress. Atomistic analysis reveals opposite stress states in the top and bottom Ti layers due to curvature-induced strain gradients. Under biaxial compression, the nanosheet buckles in a dome-like shape, whereas shear loads produce elliptical deflection modes. The presented findings may stimulate future studies on MXene morphological transformations, such as the development of nanotube, nanoscroll, and folded architectures.
arXiv:2512.22093v2 Announce Type: replace-cross
Abstract: We present an interacting model of neural network dynamics that incorporates key biological features, including multiple forms of inhibitory interactions. We develop a hierarchy of analytical mean-field approximations to characterize nonequilibrium phase transitions between ordered, disordered, and chaotic regimes, complemented by a detailed stability analysis. We show that inhibition generically enhances the stability of network dynamics. The model is consistent with the quasi-criticality hypothesis, exhibiting regions of maximal dynamical susceptibility and mutual information, modulated by the strength of external stimuli. We further demonstrate that, at the mean-field level, the critical transition belongs to the mean-field directed percolation universality class, in agreement with prior experimental and theoretical studies. More broadly, our framework may offer insights into neurological disorders, with the unstable regime exhibiting chaotic dynamics that may be associated with epileptic seizures.
arXiv:2601.04082v3 Announce Type: replace-cross
Abstract: Aluminum (Al) remains the central material for superconducting qubits, and considerable effort has been devoted to optimizing its deposition and patterning for quantum devices. However, post-processing strategies focused on oxide removal of niobium (Nb) and tantalum (Ta) -based resonators using buffered oxide etch (BOE), which can not be used for Al. This challenge becomes particularly relevant for industry-scale fabrication with multi-chip bonding, where delays between sample preparation and cooldown require surface treatments that preserve low dielectric loss during extended exposure to ambient conditions. In this work, we investigate surface modification approaches for Al resonators subjected to a 24-hour delay prior to cryogenic measurement. Passivation using self-limiting oxygen and fluorine chemistries was evaluated utilizing different plasma processes. Remote oxygen plasma treatment reduced dielectric losses, in contrast to direct oxygen plasma. A fluorine-based plasma process was developed that passivated the Al surface for subsequent BOE treatment. However, the fluorine content in the surface resulted in higher loss, identifying fluorine as an unsuitable passivation material for Al resonators. Above all, selective oxide removal using HF (hydrogen fluoride) vapor and phosphoric acid yielded median dielectric losses as low as $\tilde{\delta}_\mathrm{LP} = 5.7 \times 10^{-7}$ ($Q_\mathrm{LP} \approx 1.7\,\mathrm{M}$) with $\tilde{\delta}_\mathrm{TLS} = 3.6 \times 10^{-7}$ ($Q_\mathrm{TLS} \approx 2.8\,\mathrm{M}$) in the single photon regime. Selective oxide removal provides a promising pathway for robust Al-based qubit fabrication, as it preserves low dielectric losses for a 24-hour delay before cooldown.
arXiv:2601.07944v2 Announce Type: replace-cross
Abstract: Since the turn of the century, approximate Bayesian inference has steadily evolved as new computational techniques have been incorporated to handle increasingly complex, large-scale predictive problems. The recent success of deep neural networks and foundation models has now given rise to a new paradigm in statistical modeling, in which Bayesian inference can be amortized through large-scale learned predictors. In amortized inference, substantial computation is required at the beginning to train a neural network, but it can subsequently produce approximate posteriors or predictions at much lower computational cost across a wide range of tasks. While the typical Bayesian inference procedures are computationally expensive due to repeated likelihood calculations and Monte Carlo steps for each new dataset, amortized inference provides a much lower computational cost at deployment.
Despite the growing popularity of amortized inference, its statistical interpretation and position within Bayesian inference remain poorly explored. In this paper, we present a statistical perspective on several major neural architectures, including feedforward networks, Deep Sets, and Transformers, and examine how they naturally support amortized Bayesian inference. We explore how these models perform structured approximation and also probabilistic reasoning in ways that yield controlled generalization error throughout a wide range of deployment scenarios, and how these properties can be harnessed for Bayesian computation. Via simulation studies, we evaluate the accuracy, robustness, and uncertainty quantification of amortized inference across varying sample sizes, varying noise distributional families, varying sparsity levels, and multimodality, highlighting its strengths and limitations.
arXiv:2607.14095v1 Announce Type: new
Abstract: Retrieval Augmented Generation (RAG) has proven to be a widely successful process at improving the quality of outputs from a Large Language Model (LLM) for wider context. However, RAG systems typically retrieve context from flat document stores, which struggles when queries require hierarchical or relational reasoning across structured knowledge. I present HG-RAG (Hierarchy-Guided RAG), a framework that performs graph-traversal over a hierarchical knowledge graph to deliver structured context to a language model. My retrieval pipeline resolves a named entity anchor from the query, then expands context upward through parent nodes, laterally through relational neighbors, and downward through child nodes when needed. I evaluate HG-RAG against a dense retrieval baseline across three world scales (18-800 nodes) with four query types: local fact, hierarchical, neighborhood, and multi-hop. Results show HG-RAG consistently outperforms the flat baseline on hierarchical, relational, and multi-hop reasoning tasks, while reducing hallucination and maintaining locality coherence.
arXiv:2607.14096v1 Announce Type: new
Abstract: In predictive modeling, the ability to explain why a model produces a given target prediction has become increasingly important [5, 10]. Black-box models do not provide a transparent description of the internal mechanisms that generate the prediction, making even accurate predictions difficult to interpret and validate. In critical contexts, predictive accuracy alone is not a sufficient validation metric if the reasons underlying model decisions remain unexplained. The IMEX (Interaction-Based Model Explanation) approach represents a methodological direction within explainable predictive modeling. IMEX is designed to identify which variables contribute most to the target prediction and which interactions among variables are significant in determining the target. The method does not impose limitations on higher-order interaction analysis, allowing the investigation of feature subsets with cardinality greater than two. Beyond the identification of feature importance, IMEX enables the exploration of interaction patterns that may be consistent with latent mechanisms influencing the outcome. Through the application of the IMEX algorithm, it is possible to construct an interpretability map of the predictions. The IMEX framework is built on two complementary metrics: Static Correlation Power (PCS), which quantifies the contribution of individual features, and Interaction Correlation Power (PCI), which captures non-additive effects among features. In the present work, the PCS component is experimentally validated through a comparison with INVASE [18] on three synthetic datasets with known structures. The results indicate that IMEX can recover relevant feature-level structures in the presence of non-linear, conditional, and multicollinear relationships between input features and prediction targets.
arXiv:2607.14097v1 Announce Type: new
Abstract: We introduce RegNetAgents, an AI-oriented multi-agent framework for structured, query-driven regulatory candidate identification across heterogeneous gene regulatory networks. The system enables unified analysis of bulk tumor and single-cell-derived ARACNe networks by integrating TCGA-derived cancer networks with large-scale single-cell regulatory networks from the GREmLN project. For a given focal gene, the framework performs dual-network classification, cancer gene filtering using OncoKB annotations, and mode-of-action (MoA) assignment for tumor-derived regulatory relationships. Candidates are ranked by evidence consistency across networks (Both, TCGA-only, GREmLN-only). The system is implemented as a multi-agent LangGraph DAG workflow, accessible through a unified Python API and Model Context Protocol (MCP) client, operating as a downstream analytical layer over precomputed regulatory networks rather than a network inference method. Across eleven breast cancer (BRCA) and twelve colorectal cancer (COAD) focal genes, RegNetAgents identifies candidate regulators significantly enriched for OncoKB-annotated cancer genes. TCGA-derived candidates show strong enrichment (Stouffer Z = 6.69 for BRCA and 6.95 for COAD), while GREmLN-derived candidates also demonstrate significant enrichment (Z = 5.51 for BRCA and 7.06 for COAD; all p < 0.0001). No enrichment is observed in housekeeping or non-driver control gene sets, supporting signal specificity. An extended module enables structured evaluation of oncogenic potential, druggability, clinical relevance, and network vulnerability, supporting end-to-end interpretation from candidate identification to biological hypothesis generation. RegNetAgents establishes an interpretable AI framework for cross-network regulatory candidate identification in cancer genomics.
arXiv:2607.14099v1 Announce Type: new
Abstract: Deploying Vision-Language Models (VLMs) in real-world settings requires not only strong visual reasoning but also stability under sustained conversational pressure. We introduce Just Keep Prompting (JKP), a multi-turn evaluation framework that measures VLM epistemic stability when users repeatedly challenge, question, or contradict a model's answer. JKP probes models for up to 10 follow-up turns using three strategies: Adversarial Negation (repeated rejection), Pure Socratic Interrogation (repeated calls to reassess certainty), and Context-Aware Socratic Summarization (reflecting the model's prior rationale back before asking for reconsideration). We evaluate GPT-4o, Gemini 2.5 Pro, and Qwen3-VL-30B on a subset of the STAR benchmark across 720 multi-turn runs. Aggregate accuracy changes modestly from Turn 0 to Turn 10, but trajectory-level analysis reveals substantial instability: correct answers regress, wrong answers recover, and many runs exhibit repeated answer flipping. Repeated prompting has bounded upside and often acts as a destabilizer rather than a reasoning aid. The effect is strongly model-dependent: Qwen3-VL-30B achieves the highest final accuracy but becomes confidently wrong under direct contradiction; Gemini 2.5 Pro is comparatively stable but token-expensive; GPT-4o is the most brittle and oscillatory. These findings reveal that multi-turn VLM evaluation captures not just additional reasoning but pressure-response profiles: how models trade off visual grounding, calibration, and conversational compliance under repeated challenge.
arXiv:2607.14102v1 Announce Type: new
Abstract: With the rapid growth of digital data, real-world applications increasingly involve hierarchical information that combines static attributes with dynamic records. Modeling such heterogeneous data in a unified and generalizable manner remains challenging. Existing approaches often rely on extensive manual design, are tightly coupled to specific data schemas, and typically process static and dynamic attributes in isolation, thereby overlooking their implicit interactions. We propose UniSAGE, a unified framework for modeling data with both static and dynamic attributes. UniSAGE constructs a global attribute graph that represents hierarchical and temporal relationships in a unified structure. To ensure representational consistency, it introduces two orthogonal parameter subspaces that jointly support static aggregation and dynamic reasoning within a shared semantic space. Building on these unified representations, UniSAGE further enables task-specific interaction between static and dynamic attributes via a lightweight hyper-structure mechanism. UniSAGE is fully automated, robust to evolving data schemas, and capable of capturing complex cross-attribute dependencies. Extensive experiments on multiple public benchmarks and a real-world financial behavior dataset demonstrate that UniSAGE consistently outperforms existing methods, achieving performance improvements of over 10% on several tasks.
arXiv:2607.14119v1 Announce Type: new
Abstract: Multi-agent LLM systems commonly decompose complex tasks into specialized roles. However, this modularity introduces a representational risk: when intermediate agents transform text across linguistic registers, they can systematically compress the semantic distinctions needed for accurate downstream decisions. We term this phenomenon semantic register compression and characterize it as an observable failure mode in multi-agent cascades. Using a three-agent pipeline (Collector-Evaluator-Decider), we quantify compression via inter-label separation in sentence-transformer embedding space. Across political fact-checking (LIAR), sentiment analysis (SST-5), and medical triage (Triagegeist), critical evaluation consistently reduces label separability by 41.7% at the Evaluator stage, while identity passthrough preserves it nearly fully. Five architectural variants causally isolate oriented semantic transformation as the primary driver. A credibility-seeking variant produces minimal geometric compression yet shifts outputs toward mostly-true, demonstrating that transformation valence controls the direction of distributional collapse independently of compression magnitude. Compression generalizes across the three domains with varying intensity: 41.7% in fact-checking, 27.2% in sentiment, and 20.0% in triage. Prompt-level regression explains 78% of the variance, with operational constraints associated with lower compression. These results demonstrate that semantic register compression is a measurable and generalizable phenomenon in multi-agent LLM systems, with implications for safety evaluation in high-stakes domains.
arXiv:2607.14123v1 Announce Type: new
Abstract: Despite the proliferation of Explainable AI (XAI) techniques -- from feature attributions to sparse autoencoders -- explanations rarely influence real-world workflows. In practice, they are often generated and discarded without guiding meaningful action. This gap reflects foundational shortcomings: research has not yet established methodologies for integrating explanations into end-to-end, human-in-the-loop systems. This position paper argues that the machine learning community must pivot from ad-hoc XAI methods toward addressing foundational & structural challenges, including unclear problem formulations, underspecified evaluation objectives, and the absence of pipelines for explanation-driven feedback. We support this claim through an analysis of recent ICML, NeurIPS, and ICLR papers and a survey of XAI practitioners, revealing recurring issues that limit cumulative progress. We conclude by outlining a practical checklist designed to shift XAI toward a more human-centered, action-oriented paradigm. By emphasizing foundational clarity over the development of ad-hoc methods, we hope to provide a roadmap for integrating explanations into actionable, feedback-driven AI systems.
arXiv:2607.14146v1 Announce Type: new
Abstract: Tokamak--stellarator hybrids seek to combine tokamak-like compactness and confinement with stellarator-like externally generated rotational transform and steady-state operation. In this work, we build on the recent tokamak--stellarator hybrid study using planar dipole-field coils (PDCs) [Yu et al., arXiv:2605.03599], in which the fixed-position, programmable coils on an axisymmetric winding surface generate flexible three-dimensional shaping fields. Using single-stage free-boundary optimization of coil currents and plasma-equilibrium parameters, we construct vacuum and finite-$\beta$ configurations. The vacuum cases show controllable external transform and magnetic well. The finite-$\beta$ cases accommodate various density, temperature, and pressure profiles, producing quasi-axisymmetric (QA) equilibria with self-consistent bootstrap current, favorable Mercier stability, and reduced demand for external current drive. Re-optimization enables $\beta$ ramp-up and access to different field-period QA branches with moderate coil-current changes. At large rotational transform, a toroidally omnigenous (TO)-like configuration exhibits more favorable infinite-$n$ ideal-ballooning behavior than a QA reference with matched profiles, even though ballooning stability is not directly optimized for. These results demonstrate that PDCs provide a flexible platform for achieving optimized finite-$\beta$ hybrid configurations.
arXiv:2607.14147v1 Announce Type: new
Abstract: Aligned language models refuse harmful requests, but a one-line prefill ("Sure, here is") strips the refusal. We ask where and how it fails. The harm representation stays intact: on the prompts the attack flips to compliance, a linear probe reads harm as high as on the refused ones (0.91-0.98), while behavioral refusal drops to chance. This holds across four models and three families (1.5-3.8B, and at 14B). Refusal is therefore a shallow, response-site computation. We localize it to an early window: a dose-matched position control shows the first half of the response suffices to break refusal, while the second half is nearly inert. Three causal probes converge on that window. Restoring the harm direction there partially re-engages refusal. Injecting the model's own refuse-state reverses the jailbreak (74%, held-out). And knocking out the early response's attention to the prefill, but not an equal attention mass elsewhere, selectively collapses the harmful continuation. A base-model control identifies the mechanism: the same knockout collapses the continuation prefill-specifically even in a non-safety-tuned base model (64% to 25% harmful content vs a matched control's 64%, replicated at 7B). So the prefill's grip is generic autoregressive conditioning, not safety-specific suppression, and "refusal restoration" is a model-dependent fallback. The dominant mechanism is passive. A small safety-specific attractor remains on top (logit-trace concentration 0.24 vs 0.03), whose active-vs-passive character we size but do not fully separate. No single direction or component is a clean handle either: the decision is decodable but distributed, and refusal tracks harm rather than scary surface. The consequence is structural: a monitor reading the untouched prompt-side representation is immune by construction, but only to response-site attacks. The mechanism is diffuse; the failure surface is local.
arXiv:2607.14149v1 Announce Type: new
Abstract: Although large language models (LLMs) have set benchmarks for zero-shot reasoning, their deployment remains cost-prohibitive and environmentally taxing. Small Language Models (SLMs) offer a sustainable alternative, but prone to errors, on tasks requiring complex, multi-hop logical grounding. We investigate a neuro-symbolic agentic framework to enhance the reasoning capabilities of SLMs, specifically Gemma 3 (1B, 4B) and Llama 3.2 (3B), using the CLUTRR kinship benchmark. Our approach transforms the SLM into a minimalist agent utilizing two specialized tool calls: extract_facts for symbolic triplet extraction and get_hint for expert reasoning via a Relational Graph Convolutional Network (RGCN). We evaluate these models across two configurations, both in an Oracle scenario with ground-truth triplets and a Realistic scenario relying on self-extracted knowledge. Our results reveal that while RGCN-derived hints provide a 1.5 - 2x performance gain over story-only baselines, the system is constrained by the extraction bottleneck and sequential deductive fragility, where early extraction errors compound over multi-hop chains. Furthermore, we identify a "distraction effect" in specific architectures where noisy, self-generated facts degrade performance despite the presence of expert hints. This work characterizes the challenges of symbolic grounding in low-resource agentic systems and provides a roadmap for iterative verification in neuro-symbolic agentic pipelines.