arXiv:2606.11747v1 Announce Type: new Abstract: White-light interferometry enables ambiguity-free localization by synthesizing interference envelopes from multiple optical wavelengths, but no analogous capability has been realized for coherent matter waves. Here we report the first experimental demonstration of multi-wavelength atom interferometry, establishing the matter-wave counterpart of white-light interferometry. By exploiting counter-propagating atomic beams as multi-wavelength matter wave sources and synthesizing interference envelopes from their spectral components, we realize inertial measurements based on envelope localization rather than conventional fringe-phase estimation. The resulting multi-scale interferometric response provides ambiguity-free operation, a well-defined rotational scale factor, and reduced sensitivity to initial phase bias. As a proof of principle, we demonstrate simultaneous dual-axis rotation and acceleration sensing and directly resolve the phase ambiguity that fundamentally limits conventional open-loop atom interferometers. We further measure the Earth's rotation with a relative error of 4.3% and a long-term stability of 93 ppm at an averaging time of 15,000 s. Our results establish multi-wavelength atom interferometry as a new paradigm for coherent matter-wave sensing, extending the principles of white-light interferometry to atom optics and opening new opportunities for inertial sensing, geodesy, precision metrology, and inertial navigation.
Science Journals
arXiv:2606.11749v1 Announce Type: new Abstract: Multimodal entity linking (MEL) is the task that consists of matching textual and visual mentions of entities in unstructured data to their corresponding entities in a knowledge base (KB). To be effective in large-scale practical settings, MEL systems must meet three objectives: high linking accuracy, computational efficiency, and storage efficiency, i.e., a compact yet efficient index of the KB. In this paper, we highlight that state-of-the-art systems fail to simultaneously satisfy these 3 requirements. To meet this three-fold objective, we propose FAST-MEL, a lightweight encoder-based MEL solution that relies on a novel and compact fixed-size vectorized representation of both the textual and visual information of each entity or mention. It matches the accuracy of the best systems but performs three orders of magnitude faster. It also consumes one order of magnitude less storage than the fastest systems.
arXiv:2604.11454v2 Announce Type: replace Abstract: The GraphAlg domain-specific language for graph algorithms enables user-defined algorithms in graph databases. In this work we show how GraphAlg is built on top of the formal MATLANG language for matrix manipulation. Starting from MATLANG, we describe the extensions to MATLANG and the syntactic sugar needed to derive GraphAlg. Furthermore, we prove that any GraphAlg program can be simulated in an extension of for-MATLANG that supports simultaneous induction.
arXiv:2606.11304v1 Announce Type: new Abstract: We introduce SPADE (SPlit And Delay Embeddings), an autoregressive transformer for sequences whose tokens carry multiple features. Rather than embedding these features jointly, SPADE embeds them independently. Delaying each feature stream relative to the previous one allows intra-token correlations to be learned by the standard self-attention mechanism. Applied to point-cloud calorimeter shower generation in the highly granular ILD detector, SPADE is competitive with the state of the art AllShowers model on photon showers, and substantially outperforms its VQ-VAE-based predecessor OmniJet-$\alpha_C$. The mechanism is applicable to any generative task with multi-feature tokens, enabling LLM-style pretraining workflows for higher-dimensional data.
arXiv:2606.11753v1 Announce Type: new Abstract: We investigate polarization-selective absorption in a visible-wavelength heterostructure consisting of a ReS$_2$ stripe grating, an $\alpha$-MoO$_3$ spacer, and an Au back-reflector using finite-difference time-domain simulations. For an optimized geometry with a grating period of 500 nm, stripe width of 250 nm, and ReS$_2$ thickness of 80 nm, the structure exhibits near-unity absorption of 99.99\% at 650.5 nm under TE-polarized illumination. The resonant field is concentrated near the outer edges of the ReS$_2$ stripe, while absorption power density is localized in the same region, consistent with a localized edge mode. The absorption response depends strongly on polarization, producing a TE--TM resonance separation of 16.2 nm. Replacing either the biaxial ReS$_2$ layer or the anisotropic $\alpha$-MoO$_3$ spacer with isotropic equivalents substantially modifies the spectral response and reduces the polarization-dependent wavelength separation. In addition, rotating the crystal orientation of the ReS$_2$/$\alpha$-MoO$_3$ stack shifts both the resonance wavelength and peak absorption without changing the device geometry. The results show that the combination of anisotropic resonator and spacer layers provides an effective means of controlling resonant absorption and polarization selectivity in van der Waals photonic structures.
arXiv:2404.02147v3 Announce Type: replace Abstract: Human-centered AI (HCAI) puts the user in the driver's seat of so-called human-centered AI-infused tools (HCAI tools): interactive software tools that amplify, augment, empower, and enhance human performance using AI models. We discuss how interactive visualization can be a key enabling technology for creating such human-centered AI tools. To validate our approach, we surveyed the existing literature on HCI, AI, and visualization and interviewed researchers in relevant fields to define the characteristics of HCAI tools. We then present several examples of HCAI tools using visualization and use the examples to extract guidelines on how interactive visualization can support future HCAI tool research and development.
arXiv:2603.05573v2 Announce Type: replace Abstract: Scalable sequence models, such as Transformer variants and structured state-space models, often trade expressivity power for sequence-level parallelism, which enables efficient training. Here we examine the bounds on error and how error scales when models operate outside of their expressivity regimes using a Lie-algebraic control perspective. Our theory formulates a correspondence between the depth of a sequence model and the tower of Lie algebra extensions. Echoing recent theoretical studies, we characterize the Lie-algebraic class of constant-depth sequence models and their corresponding expressivity bounds. Furthermore, we analytically derive an approximation error bound and show that error diminishes exponentially as the depth increases, consistent with the strong empirical performance of these models. We validate our theoretical predictions using experiments on symbolic word and continuous-valued state-tracking problems.
arXiv:2603.06910v2 Announce Type: replace Abstract: Multilingual large language models (LLMs) are increasingly used in socially sensitive mental health contexts, including support chatbots, screening, and content moderation. This raises a reliability question: do semantically equivalent mental health inputs elicit comparable evaluations across languages, or systematic shifts consistent with language-associated social and cultural contexts? We examine this question in an English-Chinese setting with GPT-4o and Qwen3-32B using a two-level framework: construct-level evaluative orientation, measured by psychometric stigma instruments, and decision-level behavior, measured by binary stigma detection and four-class depression severity classification. Across instruments and models, Chinese prompts elicit higher stigma-related scores than English prompts. At the decision level, Chinese prompts reduce sensitivity to stigmatizing content and produce more conservative depression severity judgments, leading to more under-estimation errors. These findings show that prompt language can shift both evaluative orientation and downstream behavior in LLM-based mental health evaluation. They highlight the need to evaluate multilingual LLMs not only for aggregate performance, but also for whether they apply comparable evaluative standards across languages in socially sensitive domains.
arXiv:2606.11353v1 Announce Type: new Abstract: We study the Threshold Group Testing (TGT) problem in the noiseless and non-adaptive setting, where the objective is to exactly recover a sparse binary vector from pooled tests, using as few tests as possible. In TGT, each test applied to a subset of items returns a positive outcome if the number of 1's (defective items) in that subset meets or exceeds a specified threshold, and has a negative outcome otherwise. We investigate how the complexity of TGT compares to that of Classical Group Testing (CGT), corresponding to the special case of the threshold equal to one, and analyse the impact of increasing the threshold on the required number of tests. Our main contribution is the derivation of a sharp information-theoretic phase transition at $c_{\mathrm{inf}}^{\mathrm{TGT}}k\log(n/k)$ (non-adaptive) tests for TGT within the constant-column test design. The threshold constant $c_{\mathrm{inf}}^{\mathrm{TGT}}$ is expressed as a function of the prevalence of defectives and the threshold value. Our upper bound is derived under an analytic assumption, and we verify that this assumption is satisfied for a threshold value of 2. The value of $c_{\mathrm{inf}}^{\mathrm{TGT}}$ reveals that TGT on the constant-column design has the same information-theoretic behaviour as CGT in the low-prevalence regime. Yet, strikingly, at higher prevalences, the threshold leads to a significant reduction in the number of tests. On the other hand, we provide evidence that when the asymptotic proportion of defective items is positive, TGT actually becomes strictly harder than CGT (excluding trivial reductions).
arXiv:2603.22934v3 Announce Type: replace Abstract: Retrieval-Augmented Generation (RAG) improves large language model applications by grounding generation in retrieved evidence, but also introduces corpus poisoning as a new attack surface. In this setting, an adversary injects or edits passages so that they enter the Top-$K$ results for target queries and influence downstream generation. Existing defences often rely on content filtering, auxiliary models, or generator-side reasoning, which complicates deployment. We propose ProGRank, a post hoc, training-free retriever-side defence for dense-retriever RAG. ProGRank stress-tests each query--passage pair under mild randomized perturbations, extracts probe gradients from a small fixed parameter subset, and derives two instability signals: representational consistency and dispersion risk. It then combines these signals with a score gate for reranking. ProGRank preserves the original passage content, requires no retraining, and supports a surrogate-based variant when the deployed retriever is unavailable. Experiments across datasets, retrievers, attacks, and retrieval-stage and end-to-end settings show that ProGRank improves robustness and maintains a favorable robustness--utility trade-off, including under adaptive evasive attacks.
arXiv:2606.11782v1 Announce Type: new Abstract: While 3D Gaussian Splatting (3DGS) achieves impressive real-time rendering, it frequently struggles to synthesize high-frequency textures, a limitation heavily exacerbated in memory-constrained and rate-distortion-optimized (RDO) pipelines. To address this, we propose a versatile 2D perceptual wrapper that enhances the rendered outputs of existing 3DGS representations in a content- and view-dependent manner. Our method leverages a lightweight synthesis network conditioned on pseudo-random Gaussian noise to synthesize perceptually plausible textures. Supervised by Wasserstein Distortion, the network learns to match local feature statistics rather than strictly enforcing pixel-wise reconstruction fidelity, effectively mitigating the blurriness inherent in standard frameworks. We demonstrate the broad applicability of our plug-and-play approach across vanilla, memory-constrained, and RDO 3DGS methods. Comprehensive subjective and objective experiments confirm that our method significantly improves over existing baselines, yielding superior perceptual quality at sharply reduced file or model sizes.
arXiv:2606.12106v1 Announce Type: new Abstract: This paper presents our solution to the 2026 SoccerNet VQA Challenge. We first develop a cost-effective data synthesis pipeline driven by a Vision-Language Model (VLM), which systematically restructures raw domain data into diverse VQA samples, including concise answers and long-form responses. Second, we propose MSUE, a multi-expert question answering architecture that employs a Large Language Model (LLM) to dynamically dispatch questions to text, image, and video experts. These experts are instantiated as a strong text baseline Gemini3-Flash, a fine-tuned Qwen3-VL, and an external knowledge base, respectively, working collaboratively to enhance VQA performance. MSUE achieves an accuracy of \textbf{0.95} on the challenge benchmark, securing third place in the leaderboard.
arXiv:2606.12113v1 Announce Type: new Abstract: Transformer-based language models for SMILES strings suffer from a locality gap: standard character-level tokenization fragments chemically meaningful motifs, forcing models to repeatedly learn local syntax at the expense of long-range dependencies. To address this without disrupting standard tokenizers, we propose MolGram, which integrates a conditional $n$-gram memory module into molecular language models. MolGram maps local string patterns to learned embeddings via scalable hash lookups and dynamically injects this regional context into hidden states. Evaluations across three tasks, including unconditional molecule generation, forward reaction prediction, and single-step retrosynthesis, show that MolGram consistently improves performance. Crucially, our analyses demonstrate that MolGram outperforms baselines with 3$\times$ more parameters, establishing explicit local pattern memory as a highly efficient inductive bias.
arXiv:2606.11794v1 Announce Type: new Abstract: Neurodegenerative diseases such as Alzheimer's disease (AD) require accurate and scalable tools for assessing disease severity, yet current clinical staging remains time-intensive and prone to variability. We propose an attention-enhanced multimodal machine learning framework with ordinal regression for automated and interpretable AD severity staging. The framework integrates T1-weighted MRI with demographic and genetic variables and compares unimodal and multimodal architectures using ordinal and non-ordinal prediction heads. Models were trained and validated using cohort-stratified splits derived from the ADNI, AIBL, and NIFD datasets. A strictly held-out test set was constructed using subjects excluded from all training, validation, preprocessing, and hyperparameter tuning procedures, with subject-level splitting employed throughout to prevent data leakage. Among unimodal approaches, the T1-weighted MRI model achieved slightly higher adjacent-stage accuracy (0.963) and agreement with clinical staging (QWK 0.444) than the tabular model (QWK 0.433). Integrating imaging, demographic, and genetic information improved overall performance. The multimodal non-ordinal baseline achieved the lowest prediction error (MAE 0.340), whereas the ordinal multimodal model achieved the highest adjacent-stage accuracy (0.970) and strongest agreement with clinical staging (QWK 0.549). These findings indicate that ordinal formulations better capture the ordered structure of the CDR scale and yield predictions more consistent with clinical staging. Explainability analyses using Grad CAM++ and SHAP demonstrated anatomically and clinically plausible model behavior, supporting transparent decision-making. Overall, attention-based multimodal learning with ordinal regression represents a robust, interpretable, and scalable approach for automated AD severity staging and AI-assisted clinical decision support.
arXiv:2606.11796v1 Announce Type: new Abstract: Uncertainty in standalone microgrid operation usually originates from mismatches between power references and forecasts. These deviations are compensated by grid-forming controlled units, which distribute the required power contribution based on their droop gains. To introduce an additional degree of flexibility, it is possible to treat droop gains as decision variables to redistribute active-power contributions according to system-level objectives. However, directly applying updated droop gain references from a supervisory layer to the primary controllers can introduce power and frequency transients. This paper investigates transition mechanisms for applying scheduled active-power droop gain changes during operation. Hard switching, rate-limited transition, first-order IIR low-pass filtering, and cubic as well as quintic S-curve transitions are compared experimentally on two parallel 15 kW grid-forming inverter units. The results show that shaping the droop gain trajectory significantly reduces transient deviations compared to hard switching. In the considered case study, the S-curve transitions provide the strongest transient mitigation, reducing the active-power overshoot from 632.7 W to approximately 115 W and limiting the frequency overshoot to about 0.003 Hz.
arXiv:2604.08447v3 Announce Type: replace Abstract: The present research applies Graph Neural-Networks (GNNs) for energy measurement and particle identification tasks for a proposed second detector at the future Electron Ion Collider (EIC). In particular, an iron-scintillator sampling calorimeter would provide neutral hadron ($K_L$ and neutron) energy measurements and identification, as well as separation of muons from hadrons. Using detector simulations, particle hits in the detector are represented as graphs, and a GNN is trained for either classification or prediction. Furthermore, we developed a parameterization of the scintillator optical photon simulation that yields a 20-fold speed up compared to the default simulation. We find that the GNN method outperforms classical methods at the same tasks, and we report projections for the energy and timing resolution, and identification accuracy of the calorimeter. We also present an integration of the GNN method into a Multi-Objective Optimization framework, enabled by an automated pipeline of data generation, GNN training, and detector performance evaluation. We utilize the optimization to quantify the tradeoffs between different performance metrics at high and low energies when changing the detector design parameters, such as the iron/scintillator thickness.
arXiv:2604.21413v2 Announce Type: replace Abstract: Enterprise data exists in many forms, such as tables, text, maps, e-mail, and CAD models, that are access-controlled and hidden behind bespoke interfaces. Current agentic AI systems delegate the entire query workflow to a frontier LLM: a single model interprets the request, selects sources or tools, integrates retrieved evidence, judges completeness, and generates an answer, with few constraints, limited use of schemas, and text as the primary representation throughout. We argue that this is an ineffective abstraction for enterprise data. Reliable agentic AI should instead require structure: a constrained query interface over each source and a table-centric integration layer driven by a query processor. We introduce RUBICON, a system that embodies this vision. RUBICON is based on two observations. First, text-to-SQL fails on real enterprise data and must be dramatically subsetted to achieve reliable results. Second, data integration across disparate corporate datasets is best performed using tables as the core abstraction rather than text-centric LLM pipelines. We evaluate RUBICON on two benchmarks: our enterprise-focused RUBICON-Bench, against agentic baselines, and SemBench, against LOTUS and Palimpzest. On RUBICON-Bench, where queries require coordination across heterogeneous enterprise sources, RUBICON achieves 100% end-to-end accuracy, while all agentic baselines, including single- and multi-agent ReAct systems, produce no correct answers. On SemBench, RUBICON surpasses both LOTUS and Palimpzest: it achieves 14.7% higher accuracy, reduces latency by 62.64%, and lowers token cost by 98.64%, demonstrating that a table-centric architecture better matches enterprise data while yielding significant efficiency gains.
arXiv:2604.23422v2 Announce Type: replace Abstract: Semiconductor quantum dots (QDs) offer outstanding quantum-optical properties, making them highly attractive for quantum information technologies. However, combining wide-range electrical tunability, efficient photon extraction, elevated-temperature operation, monolithic silicon integration, and telecom-wavelength compatibility remains a major challenge. Here, we demonstrate electrically contacted circular Bragg grating (eCBG) resonators incorporating InGaAs QDs directly grown on silicon, enabling bright single-photon emission in the telecom O-band. Deterministic electron-beam lithography and a ridge-based vertical p--i--n diode architecture enable precise device integration and electrical control of individual emitters. The QD--eCBGs exhibit a quantum-confined Stark shift of approximately 16 nm (11 meV) at 4 K, representing a record for QDs embedded in nanophotonic structures at telecom wavelengths. This is achieved alongside a photon extraction efficiency of $(21.7 \pm 3.0)\%$ into the first lens, while maintaining excellent radiative properties and high single-photon purity, with $g^{(2)}(0)=0.0078 \pm 0.0012$ below saturation and $g^{(2)}(0)=0.0183 \pm 0.0021$ at saturation under pulsed excitation. Robust antibunching persists up to 77 K, with $g^{(2)}(0)=0.0663 \pm 0.0056$, enabling operation with liquid-nitrogen or compact Stirling cryocoolers. Furthermore, spatially separated QD--eCBGs can be electrically tuned into spectral resonance without degrading photon statistics. These results establish a silicon-compatible, electrically addressable telecom O-band quantum light platform combining wide spectral tunability, high single-photon purity, and elevated-temperature operation, providing a scalable route toward practical photonic quantum networks.
arXiv:2605.02411v2 Announce Type: replace Abstract: A semantic gap separates how users describe tasks from how tools are documented. As API ecosystems scale to tens of thousands of endpoints, static retrieval from the initial query alone cannot bridge this gap: the agent's understanding of what it needs evolves during execution, but its tool set does not. We identify this retrieval interface, not planning, as the binding constraint on end-to-end agent performance, and introduce FitText, a training-free framework that makes retrieval dynamic by embedding it directly in the agent's reasoning loop. FitText treats retrieval as test-time evolution of hypotheses: the agent generates natural-language pseudo-tool descriptions (revisable beliefs about the tool it needs), refines them iteratively using retrieval feedback, and explores diverse alternatives through stochastic generation. Memetic Retrieval adds evolutionary selection pressure over candidate descriptions, guided by a tool memory that avoids redundant search. On ToolRet (three domains), FitText's reformulation strategies improve NDCG@5 by 2.7 to 10.6 points over static query retrieval across all base models; on StableToolBench (16,464 APIs) with GPT-5.4-mini, Memetic reaches an 84.3% pooled pass rate, a 26.7-point absolute gain over static query retrieval.
arXiv:2605.06275v2 Announce Type: replace Abstract: Fluid antenna systems (FAS) exploit antenna position reconfigurability to unlock massive spatial diversity within compact form factors, making them a promising enabler for 6G user terminals (UTs). However, practical port switching incurs latency and signaling overhead, which can be particularly detrimental to hyper-reliable low-latency communications (HRLLC) under finite blocklength operation. This paper investigates FASenabled HRLLC by explicitly capturing the coupled effects of spatial correlation, port switching delay, and finite blocklength coding. We derive exact closed-form expressions for the average block error rate (BLER) and average achievable rate over spatially correlated fading channels. The resulting analysis reveals a fundamental design trade-off: increasing the number of ports improves diversity but linearly reduces the effective blocklength, thereby intensifying finite-blocklength penalties. A key theoretical contribution is a rigorous proof that reliability, achievable rate, and energy efficiency are strictly unimodal in the port dimension, ensuring a unique optimal port configuration. Furthermore, we characterize an explicit switching-delay threshold that separates regimes where FAS yields net gains over fixed-position antenna (FPA) systems. Numerical results validate the analysis and show that substantial HRLLC performance gains are achievable when the switching latency remains below the derived bound.
arXiv:2605.09059v2 Announce Type: replace Abstract: Code generation is one of the tasks for which the use of Large Language Models is widely adopted and highly successful. Given this popularity, there are many benchmarks dedicated to code generation that can help select the best model. However, they primarily focus on measuring solution correctness, leaving other aspects, such as code quality and usability, behind. This paper aims to describe a custom tree-fold evaluation methodology for code generated by Large Language Models that bridges this gap. The methodology includes a dedicated correctness benchmark based on a complex multi-level computer science project, code quality verification, and a survey of developers' opinions on generated code samples gathered through a structured code-review process. The proposed methodology's usage and usefulness are demonstrated by evaluating and comparing three general-purpose Large Language Models: GPT-4.1, DeepSeek-V3-0324, and Claude Opus 4. The results show that reviews gathered from developers can yield many new findings, especially those related to the code being in a production-ready state, that would not be possible to obtain using the standard correctness-focused benchmark approach.
arXiv:2605.12655v3 Announce Type: replace Abstract: Multi-agent reinforcement learning (MARL) in real-world use cases may need to adapt to external natural language instructions that interrupt ongoing behavior and conflict with long-horizon objectives. However, conditioning rewards on instructions introduces a fundamental failure mode as Bellman updates couple value estimates across instruction contexts, leading to inconsistent values when instructions interrupt macro-actions. We propose Macro-Action Value Correction for Instruction Compliance (MAVIC), which corrects Bellman backups at instruction boundaries by correcting the incoming instruction objective and restoring the continuation value under the current objective. Unlike reward shaping, MAVIC modifies the bootstrapping target itself, enabling consistent value estimation under stochastic instruction switching within a unified policy. We provide theoretical analysis and an actor-critic implementation, and show that MAVIC achieves high instruction compliance while preserving base task performance in increasingly complex cooperative multi-agent environments.
arXiv:2606.11824v1 Announce Type: new Abstract: The rapid evolution of cloud computing has resulted in the adoption of hybrid deployments that blend Infrastructure-as-a-Service (IaaS) and Function-as-a-Service (FaaS) service models to optimize resource utilization, scalability, and operational efficiency. This paper presents a comprehensive study and practical implementation of a metrics-driven approach for migrating microservices from a traditional IaaS service model to a hybrid IaaS + FaaS model, using two microservice applications as case studies. The research develops an automated framework to analyze service-level performance metrics to identify microservices that are best suited for serverless execution. The findings of our research highlight the benefits and limitations of different cloud service models and provide a scalable and replicable automated methodology for optimized deployment of cloud-native applications.
arXiv:2605.13674v2 Announce Type: replace Abstract: Weakly supervised semantic segmentation (WSSS) trains dense pixel-level segmentation models from partial or coarse annotations such as bounding boxes, scribbles, or image-level tags. While recent work leverages foundation models such as the Segment Anything Model (SAM) to generate pseudo-labels, these approaches typically depend on heuristic prompt choices and offer limited ways to incorporate prior knowledge or heterogeneous labels. We address this gap by taking a neurosymbolic perspective: integrating differentiable fuzzy logic with deep segmentation models. Weak annotations and domain-specific priors are unified as continuous logical constraints that fine-tune SAM under weak supervision. The refined foundation model then produces improved pseudo-labels, from which we train a second-stage prompt-free segmentation model. Experiments on Pascal VOC 2012 and the REFUGE2 optic disc/cup segmentation dataset show that our logic-guided fine-tuning yields higher-quality pseudo-labels, leading to state-of-the-art segmentation accuracy that often exceeds densely supervised baselines.
arXiv:2606.11831v1 Announce Type: new Abstract: Neural relational inference (NRI) methods discover interaction graphs from trajectories through variational reasoning on discrete potential edges. However, these methods typically rely on oversimplified, factorized graph priors. Such priors, typically nearing uniform distributions, treat edges as independent entities. This systemic misalignment does not match the real-world systems and yields diffuse and indecisive edge posteriors limiting the reliability of structural discovery. To address this, we propose \textit{Diff-prior}, a diffusion-parameterized adaptive prior used to calibrate latent graph distribution rather than generate graphs. Our core insight is to reframe prior integration as a learnable denoising-style calibration that organizes scattered, uncertain edge posteriors into a more reliable overall structure which can be trained by the diffusion model. Diff-prior learns an adaptive structure prior that performs structured calibration on the edge posteriors during inference, guiding it towards a distribution closer to the underlying structure. The diff-prior operates before structural sampling and acts as a denoising calibrator directly on the encoder edge distribution, which provides a generic training paradigm over structured variables. Experiments on standard benchmarks validated our framework, and the results indicate that Diff-prior improves the performance of structure inference and generates more decisive edge posteriors across multiple NRI-family architectures. The code is available on https://github.com/Hardy158118/Diffprior.