arXiv:2607.10470v1 Announce Type: new
Abstract: Real-world deployment of vision models to broadly benefit society is arguably a main research objective. In optical flow, however, the difficulty to obtain the ground truth has focused research mainly on synthetic data and domain-specific benchmarks. Here, we investigate the severity of this mismatch. We study how well modern optical flow estimation models generalise to real-world video and question if accuracy on synthetic benchmark proxies actually predicts accuracy on real-world optical flow. To address this, we build a real-world evaluation benchmark and evaluate the real-world generalisability of a broad set of recent optical flow models using standard checkpoints. Our benchmark contains 8,204 frame pairs across TAP-Flow, Slow Flow, and our own dataset FlowFactor. FlowFactor is a manually annotated real-world benchmark of 1,000 HD frame pairs organised into four confounding factors: large displacements, repetitive textures, occlusions, and lighting variation. Each setting mainly varies only one factor, enabling diagnostic, confounder-specific analysis. Using FlowFactor, we reveal that performance on varying lighting and large displacements correlates most strongly with real-world accuracy, and that improvements on large-motion regimes can trade off against robustness in small-motion, stationary scenes. Our experiments show that progress on Sintel, KITTI and Spring only weakly predicts accuracy on real-world data, highlighting the need for a broad real-world optical flow benchmark. Interestingly, scaling up the amount of training data does not necessarily resolve the gap, calling for new innovative research instead of simply scaling data and compute.
Science Journals
arXiv:2607.09665v1 Announce Type: new
Abstract: Prompt wrappers often differ only in formatting, yet they can change model scores enough to flip leaderboard conclusions. We study this variance under a token-controlled protocol and introduce two complementary metrics: the Format Sensitivity Index (FSI), the accuracy range induced by wrapper choice, and the Parseability Sensitivity Index (PSI), the corresponding range in answer parseability. Across 140,000 OpenRouter generations spanning 7 QA tasks, 5 wrapper families, and 4 instruct models from 7B to 72B parameters, we find that mean FSI varies by over 30x across models and is largely explained by compliance failures. A fixed-effects regression shows that parseability remains a strong predictor of accuracy even after controlling for task, model, and wrapper. We argue that reporting accuracy without wrapper variance and compliance is statistically fragile, and we give practical recommendations for both benchmarking and structured-output deployments.
arXiv:2607.11203v1 Announce Type: new
Abstract: In online graph exploration, introduced by Kalyanasundaram and Pruhs (1994), an agent must visit all vertices of an initially unknown weighted graph and return to its starting position, while the graph is revealed only locally at visited vertices. Although the problem has attracted considerable attention, previous work has focused exclusively on deterministic algorithms. Randomized strategies are often substantially harder to analyze because of a fundamental challenge inherent to exploration. In this work, we give the first positive result showing that randomization can improve competitive guarantees in online graph exploration. To this end, we focus on cycles, a simple graph class which nevertheless captures a key difficulty of online exploration.
Our main contribution is \(\textsc{RandHeavyTest}\), a randomized algorithm for online exploration of cycles whose competitive ratio we prove to be at most 1.315. This establishes a strict separation from the deterministic setting, where the optimal competitive ratio is $\thickapprox 1.366$, and thus gives the first provable advantage of randomization in online graph exploration. A key step towards this result is a new, simplified optimal deterministic algorithm, \(\textsc{HeavyTest}\), whose formulation naturally suggests the randomized variant. We complement our upper bounds with lower bounds of 1.115 for arbitrary randomized algorithms and 1.207 for the natural class of so-called forward-greedy algorithms, which includes \(\textsc{RandHeavyTest}\).
arXiv:2607.10476v1 Announce Type: new
Abstract: Large language models (LLMs) have emerged as a powerful tool for retrieving knowledge through seamless, human-like interactions. Despite their advanced text generation capabilities, LLMs exhibit hallucination tendencies, where they generate factually incorrect statements and fabricate knowledge, undermining their reliability and trustworthiness. Multiple studies have explored methods to evaluate LLM uncertainty and detect hallucinations. However, existing approaches are often probabilistic and computationally expensive, limiting their practical applicability. In this paper, we introduce diversion decoding, a novel method for developing an LLM uncertainty heuristic by actively challenging model-generated responses during the decoding phase. Through diversion decoding, we extract features that capture the LLM's resistance to produce alternative answers and utilize these features to train a machine-learning model to develop a heuristic measure of the LLM's uncertainty. Our experimental results demonstrate that diversion decoding outperforms existing methods with significantly lower computational complexity, making it an efficient and robust solution for evaluating hallucination detection.
arXiv:2607.10546v1 Announce Type: new
Abstract: Discovering governing partial differential equations (PDEs) from noisy observational data is a fundamental challenge in scientific machine learning. Traditional symbolic regression (SR) methods often struggle to identify accurate equations within vast combinatorial search spaces, largely due to their inability to incorporate essential domain-specific prior knowledge. Furthermore, reliance on pointwise evaluations and discrete finite differences inherently amplifies high-frequency noise, creating deceptive fitness landscapes that derail the optimization process. To resolve these bottlenecks, we propose LLM-PDESR, a framework that integrates the structural hypothesis generation of Large Language Models (LLMs) with a mathematically rigorous evaluation environment. By employing C^4-continuous quintic splines for robust differentiation and subdomain weighted residuals as natural low-pass filters, our approach effectively mitigates the fitness landscape distortion that plagues existing methods. A Pareto-driven feedback loop then enables the LLM to iteratively refine candidate equations, balancing predictive accuracy with structural parsimony. We evaluate LLM-PDESR on 23 canonical PDEs and five structurally novel equations (including a multivariate system) specifically designed to preclude dataset memorization and test true discovery capabilities. Demonstrating real-world applicability, the framework successfully extracts a consistent structural skeleton for an interpretable 1D dynamical surrogate (1D-CACE) directly from noisy ERA5 reanalysis data. Extensive experiments and out-of-distribution testing confirm that LLM-PDESR significantly outperforms state-of-the-art methodologies in structural recovery, noise resilience, and the avoidance of spurious complexity and equation bloat.
arXiv:2607.09685v1 Announce Type: new
Abstract: Policy models often assume that the relationship between a policy instrument and its outcome remains stable across institutional conditions. In adaptive socio-technical systems this assumption may fail: regulatory change can alter incentives, agents can respond strategically, and the mapping from policy variables to aggregate outcomes can change. This paper studies such regime change as a transfer-learning problem in adaptive multi-agent systems. A policy regime is represented as a learning problem induced by an observable input distribution and a target function mapping policy variables to outcomes. We compare a blank-slate learner that searches a flexible hypothesis class in the new regime with a transfer learner whose effective hypothesis class is restricted by structural knowledge from the previous regime. Transfer is beneficial when this restriction preserves the new target function while reducing effective complexity; it is harmful when the restriction excludes the new target and creates misspecification. A stylized emissions-regulation experimental environment and a dynamic ABM robustness experiment support the claim. When the target regime preserves an affine monotone tax-emissions relation, transfer improves empirical small-sample performance. When the target regime introduces a threshold break, the same transferred structure produces negative transfer: held-out error remains high, online prediction generates more mistakes, and repeated online streams show larger cumulative and final-window error under misspecification. The contribution is methodological: previous regulatory experience should be reused when it captures stable structural invariants, but treated cautiously when policy change alters the policy-outcome relationship.
arXiv:2607.10364v1 Announce Type: new
Abstract: In radiation transfer simulations, an M1 method achieves substantial computational savings by replacing the full angular transport equation with a low-order moment system. Because this reduced system is not closed, a closure model is required to represent the unknown higher-order moments using lower-order moments. While machine learning (ML)-based closures can improve accuracy beyond classical analytic closures, unconstrained learned closures may produce non-real characteristic speeds and consequently cause numerical solver breakdown. To guarantee real eigenvalues of the Jacobian associated with ML closures, we propose a hyperbolic neural closure for the M1 radiative transfer system. Rather than directly predicting closure terms, we parameterize the Jacobian through two neural networks: (i) a symmetric matrix network and (ii) a strictly convex entropy network whose Hessian defines a positive definite symmetrizer. These components are combined to yield a Jacobian that is similar to a symmetric matrix, thereby ensuring real eigenvalues. The closure is then reconstructed by numerical integration of the learned Jacobian field along a prescribed integration path. Numerical experiments show that the proposed closure not only achieves higher closure accuracy than classical analytic closures, but also improves solution accuracy and remains stable in discontinuous Galerkin simulations for radiative transfer problems.
arXiv:2506.20893v5 Announce Type: replace
Abstract: In this paper, we reveal a significant shortcoming in class unlearning evaluations: overlooking the underlying class geometry can cause information leakage about the forgotten class. We further propose a simple unlearning strategy to mitigate this issue. We introduce Class Membership Inference Attack (CMIA) that uses the probabilities assigned by the model to neighboring classes to detect unlearned samples. We find that existing unlearning methods are vulnerable to CMIA across multiple datasets. We then propose a new fine-tuning objective that mitigates this privacy leakage by approximating, for forget-class inputs, the distribution over the remaining classes that a retrained-from-scratch model would produce. To construct this approximation, we estimate inter-class similarity and tilt the target model's distribution accordingly. The resulting Tilted REWeighting (TREW) distribution serves as the desired distribution during fine-tuning. We also show that across multiple benchmarks, TREW matches or surpasses existing unlearning methods on prior unlearning metrics. More specifically, on CIFAR-10, it reduces the gap with retrained models by 19% and 46% for U-LiRA and CMIA scores, accordingly, compared to the SOTA method for each category. Our code is publicly available at https://github.com/CrowdDynamicsLab/ICML_TREW.
arXiv:2607.09699v1 Announce Type: new
Abstract: The Bekenstein--Hawking entropy $S = k_B A/4\lp^2$ contains a coefficient $1/4$ that has been calculated by many approaches to quantum gravity. We show that this factor arises from the causal geometry of any null boundary in a four-dimensional Lorentzian spacetime. A heuristic accessibility argument is supported by a purely geometric derivation using the canonical symplectic structure on the cotangent bundle and the causal splitting of the space of null rays. No gravitational field equations are invoked.
arXiv:2607.09709v1 Announce Type: new
Abstract: Post-training a code generator against a learned judge can optimize proxy features that raise the score without improving the artifact. We study the opposite signal: a deterministic, judge-free, ungameable filter -- whether a generated project launches cleanly under a headless engine (strict-launch). Under this gate, rejection-sampling self-distillation compounds out-of-family generalization. On GameCraft-Bench (mapping a natural-language brief to a complete Godot project), a 14B model (Qwen3-14B+LoRA) distilled under strict-launch raises clean generation on four unseen game families from 8.8% to 42.2% per-candidate and best-of-K coverage from 18/25 to 25/25 (the gold ceiling) over three rounds, each a significant gain (p=0.0019, p<1e-4, p<1e-4). The gain is not from merely adding data: an exactly-matched gold-duplication control regresses below the base model (5.6% vs. 8.8%, p=0.019), while a count-matched decomposition splits the round-1-to-2 jump into comparable quality (+8.8pp) and quantity (+8.5pp) channels. Most directly, rerunning the loop with only the filter swapped -- the lenient BUILD check, which passes 99.9% of generations, in place of the launch gate -- erases the gain entirely (back to base, p=1e-3 vs. the launch-gated round), isolating verifier precision rather than the optimizer. A second ungameable signal, headless execution grounding, rises monotonically across rounds and yields far more grounded candidates than gold-duplication at a matched budget (16 vs. 5), confirming the gains are functional, not launch-but-empty. Game generation is a verifiable testbed for one lesson: the verifier is the curriculum -- what it certifies is what the model learns.
arXiv:2507.07565v5 Announce Type: replace
Abstract: This work studies privacy-preserving federated learning (ppFL) under unreliable communication. In ppFL, zero-sum privacy noises enables privacy protection without sacrificing model accuracy, effectively overcoming the privacy-utility trade-off. However, in practice, unreliable communication can randomly disrupt the coordination of zero-sum noises, leading to aggregation errors and unpredictable partial participation, which severely harm the model accuracy and learning performance. To overcome these challenges, we propose a robust coding-enforced structured secure aggregation method, termed secure cooperative gradient coding (SecCoGC), which enables exact reconstruction of the global model under unreliable communication while allowing for arbitrarily strong privacy preservation. In this paper, a complete problem formulation and constructions of real-field zero-sum privacy noise are presented, and fairness is introduced as a privacy metric. Privacy across all protocol layers in SecCoGC is evaluated, accounting for the correlation among privacy noises and their linear combination under unreliable communication. Moreover, a distinct convergence analysis for the FL algorithm with a binary outcome for global model recovery is provided. Experimental results demonstrate that SecCoGC achieves strong resilience to unreliable communication while maintaining varying levels of privacy preservation, yielding test accuracy improvements of up to 20%-70% over existing benchmark methods.
arXiv:2507.08825v2 Announce Type: replace
Abstract: How can one design complex systems capable of learning for a given functionality? In the context of ultrafast laser-surface interaction, we unravel the nature of learning schemes tied to the emergence of complexity in dissipative structures. The progressive development of learning mechanisms, from direct information storage to the development of smart surfaces, originates from the network of curvatures formed in the unstable fluid under thermoconvective instability, which is subsequently quenched and resolidified. Under pulsed laser irradiation, non-equilibrium dynamics generate intricate nanoscale patterns, unveiling adaptive process mechanisms. We demonstrate that the imprints left by light act as a form of structural memory, encoding not only local effects directed by laser field polarization but also a cooperative strategy of reliefs that dynamically adjust surface morphology to optimize light capture. By investigating how apparent complexity and optical response are intricately intertwined, shaping one another, we establish a framework that draws parallels between material adaptation and learning dynamics observed in biological systems.
arXiv:2507.11068v2 Announce Type: replace
Abstract: The Active Flux method is a numerical method for conservation laws using a combination of cell averages and point values as independent degrees of freedom, based on ideas from finite volumes and finite differences. This unusual mix has been shown to work well in many situations. We expand the theoretical justifications of the Active Flux method by analyzing it from the point of view of summation-by-parts (SBP) operators, which are routinely used to analyze finite difference, finite volume, and finite element schemes. We investigate in what type of setting the Active Flux method can be formulated using classical or degenerate SBP operators, yielding a first and novel approach for showing the energy stability of the Active Flux method. We present the analysis for the one-dimensional scalar linear advection equation with periodic boundary conditions on a uniform grid.
arXiv:2607.09738v1 Announce Type: new
Abstract: The standard Dirac Lagrangian is linear in the field derivatives and therefore has a vanishing Hessian. We identify the minimal null deformations of this Lagrangian that make the covariant Legendre map locally invertible. Imposing global phase invariance, reality, proper Poincar\'e covariance, and absence of external background tensors leaves the two-parameter spinorial bivector $\mathcal E^{\mu\nu}=\ell\sigma^{\mu\nu} +\ell_5{}^\star\sigma^{\mu\nu}$, where the star denotes the Hodge dual and $\ell^2+\ell_5^2\neq0$. This extends the analysis of \cite{struckmeier2024pauli} by a parity-odd term. After minimal $U(1)$ gauging, this free null term is no longer variationally trivial and induces magnetic- and electric-dipole Pauli operators, together with an identically conserved dipole current. These dipole terms make the species-dependent regularization lengths directly constrainable by precision moment measurements. Metric-affine gauging in a Lorentz-spinor prescription then produces spin-curvature, torsion, and nonmetricity corrections to the Dirac operator. Applying the Velo-Zwanziger criterion, we exclude all nonzero pure axial, pure trace, and mixed axial-trace torsion backgrounds, as well as nonzero pure Weyl and second-trace nonmetricity backgrounds. The combined trace-vector nonmetricity sector is not excluded only when the effective trace vector vanishes. Tensor torsion and general tracefree nonmetricity remain unclassified without further algebraic assumptions, while the Levi-Civita limit preserves the metric light cone and leaves only lower-order curvature-dependent mass terms. Thus, gauging a Legendre-regular Dirac representative turns a free variational ambiguity into observable couplings, while non-Riemannian sectors are sharply constrained by causality bounds.
arXiv:2607.10096v1 Announce Type: new
Abstract: Embedding-based retrieval (EBR) is foundational to large-scale e-commerce search, yet its effectiveness is often constrained by the quality of training signals and the representational capacity of the encoder. Standard dual-encoders suffer from a training-inference gap: they are optimized on narrow candidate pools but must discriminate against hundreds of millions of items during inference. Furthermore, while transitioning to higher-capacity backbones can mitigate this gap, simply replacing a mature model can lead to inconsistent retrieval behavior and a loss of the domain-specific knowledge established in previous iterations. In this paper, we present a unified pipeline deployed at Walmart that addresses both signal quality and model evolution. Our contributions are two-fold: (1) Hybrid Hard Negative Mining: We integrate Online Cross-Batch Sampling to increase negative diversity by an order of magnitude and Hybrid Offline Mining, which combines cross-encoder predictions with metadata heuristics to identify nuanced mismatches. (2) Legacy-Aware Distillation: We transition from DistilBERT to a higher-capacity GTE-base encoder. To ensure a smooth and superior transition, we introduce a Warm-Start Distillation technique that transfers domain-specific expertise from the legacy model to the new backbone. Validated through extensive offline experiments and online A/B testing, the proposed pipeline is deployed in live production, delivering a +7.34% improvement in NDCG@5 and a +0.50% lift in gross revenue.
arXiv:2508.17728v3 Announce Type: replace
Abstract: Cervical cancer remains a significant global health concern and a leading cause of cancer-related deaths among women. Early detection through Pap smear tests is essential to reduce mortality rates; however, the manual examination is time consuming and prone to human error. This study proposes a deep learning framework that integrates U-Net for segmentation and a classification model to enhance diagnostic performance. The Herlev Pap Smear Dataset, a publicly available cervical cell dataset, was utilized for training and evaluation. The impact of segmentation on classification performance was evaluated by comparing the model trained on segmented images and another trained on non-segmented images. Experimental results showed that the use of segmented images marginally improved the model performance on precision (about 0.41 percent higher) and F1-score (about 1.30 percent higher), which suggests a slightly more balanced classification performance. While segmentation helps in feature extraction, the results showed that its impact on classification performance appears to be limited. The proposed framework offers a supplemental tool for clinical applications, which may aid pathologists in early diagnosis.
arXiv:2508.19473v2 Announce Type: replace
Abstract: This paper shows a polynomial-time algorithm that, given a general matroid $M_1$ and $k-1$ partition matroids $ M_2, \ldots, M_k$, produces a coloring of the intersection $M = \cap_{i=1}^k M_i$ using at most $1+\sum_{i=1}^k \left(\chi(M_i) -1\right)$ colors. This is the first polynomial-time $O(k)$-approximation algorithm for matroid intersection coloring where one of the matroids may be a general matroid. Leveraging the fact that most of the standard combinatorial matroids reduce to partition matroids at a loss of a factor of two in the chromatic number, this algorithm also yields a polynomial-time $O(k)$-approximation algorithm for matroid intersection coloring in the case where each of the matroids $ M_2, \ldots, M_k$ are one of these standard combinatorial types. Even when $k = 2$, the previous best-known approximation ratio was $O(\log n)$ via a reduction to Set Cover.
arXiv:2509.24256v2 Announce Type: replace
Abstract: The pretrain-transfer paradigm, which underpins the success of large language models (LLMs), has demonstrated the immense power of creating foundation models that learn generalizable representations from vast datasets. However, extending this paradigm to Operations Research (OR) problems on graph structures remains challenging due to the fundamental conflict between the statistical flexibility of language and the strict combinatorial constraints of graphs. To bridge this gap, we introduce the Graph Foundation Model (GFM), the first framework capable of solving all distance-based optimization problems on graph structures. By introducing the LLM-like self-supervised pre-training paradigm on the paths generated from random walks in the graph, GFM is compelled to internalize the graph's complex topological and combinatorial rules, where the connectivity of the structure itself can be treated as the supervisory signal. Unlike existing neural methods that learn complex and task-specific solving policies, our approach leverages the pre-trained GFM as a foundational model of the graph's intrinsic structure, which in turn enables a simple generative heuristic to tackle a diverse range of optimization challenges effectively. Comprehensive experiments on networks ranging from 20 to 893 nodes demonstrate that GFM achieves competitive performance against specialized solvers across a variety of distinct optimization task classes, while maintaining significantly faster inference times. Our work establishes a new paradigm of adapting the pretrain-transfer framework to graph optimization, opening the door for applying foundation model innovations to OR.
arXiv:2510.06717v2 Announce Type: replace
Abstract: Large language models (LLMs) have been widely applied to knowledge-driven decision-making for automated vehicles due to their strong generalization and reasoning capabilities. However, the safety of the resulting decisions cannot be ensured due to possible hallucinations and the lack of integrated vehicle dynamics. To address this issue, we propose SanDRA, the first safe large-language-model-based decision making framework for automated vehicles using reachability analysis. Our approach starts with a comprehensive description of the driving scenario to prompt LLMs to generate and rank feasible driving actions. These actions are translated into temporal logic formulas that incorporate formalized traffic rules, and are subsequently integrated into reachability analysis to eliminate unsafe actions. We validate our approach in both open-loop and closed-loop driving environments using off-the-shelf and finetuned LLMs, showing that it can provide provably safe and, where possible, legally compliant driving actions, even under high-density traffic conditions. To ensure transparency and facilitate future research, all code and experimental setups are publicly available at github.com/CommonRoad/SanDRA.
arXiv:2510.15574v2 Announce Type: replace
Abstract: In this article, we design and analyze a hybrid high-order (HHO) finite element approximation for the solution of a nonlocal nonlinear problem of Kirchhoff type. The HHO method involves arbitrary-order polynomial approximations on structured and unstructured polytopal meshes. We establish the existence of a unique discrete solution to the nonlocal nonlinear discrete problem. We derive an optimal-order error estimate in the discrete energy norm. The discrete system is solved using Newton's iterations on the sparse matrix system. We perform numerical tests to substantiate the theoretical results.
arXiv:2510.17500v2 Announce Type: replace
Abstract: In this paper, a finite volume approximation scheme is used to solve a nonlocal macroscopic material flow model in two space dimensions, accounting for the presence of boundaries in the nonlocal terms. Based on a previous result for the scalar case, we extend the setting to a system of heterogeneous material on bounded domains. We prove the convergence of the approximate solutions constructed using the Roe scheme with dimensiona splitting, where the major challenge lies in the treatment of the discontinuity occurring in the flux function. Numerical tests show a good agreement with microscopic simulations.
arXiv:2510.21859v3 Announce Type: replace
Abstract: Electromagnetic (EM) methods, owing to their efficiency and non-invasive nature, have become one of the most widely used techniques in geological exploration. Nevertheless, data processing for these methods remains highly time-consuming and labor-intensive. With the remarkable success of deep learning, applying such techniques to EM methods has emerged as a promising research direction to overcome the limitations of conventional approaches. The effectiveness of deep learning methods depends heavily on the quality of datasets, which directly influences model performance and generalization ability. Existing application studies often construct datasets from random one-dimensional or structurally simple threedimensional (3D) models, which fail to represent the complexity of real geological environments. Furthermore, the absence of standardized, publicly available 3D geoelectric datasets continues to hinder progress in deep learning based EM exploration. To address these limitations, we present OpenEM, a large-scale, multi-structural 3D geoelectric dataset that encompasses a broad range of geologically plausible subsurface structures. OpenEM consists of nine categories of geoelectric models, spanning from simple configurations with anomalous bodies in half-space to more complex structures such as flat layers, folded layers, flat faults, curved faults and their corresponding variants with anomalous bodies. In addition, we provide a 3D model generator that enables fully controllable 3D model construction, allowing flexible and extensible augmentation of OpenEM. OpenEM provides a unified, comprehensive, and large-scale dataset for common EM exploration systems to accelerate the application of deep learning in electromagnetic methods. The complete dataset and 3D model generator is publicly available at https://doi.org/10.5281/zenodo.17141981.
arXiv:2607.10265v1 Announce Type: new
Abstract: Temporal graph questions require reliable handling of time, identifiers, and arithmetic. Large language model (LLM) agents often fail on these tasks, especially when a graph records both ordinary evolution and later corrections. We present TGMS, a bi-temporal property graph management system that exposes thirteen verified temporal operators as agent tools. Each operator is typed, deterministic, bounded, cost-guarded, and bi-temporal by default. The LLM plans operator calls and writes the final response, while the system performs all graph computation. Numeric, entity, ordering, and pattern claims are checked against the content-addressed execution trace.
TGMS separates valid time from transaction time. It can therefore answer belief-state questions such as ``as of transaction time $T$, what did the system believe?'' Standard latest-state snapshots and retrieval pipelines do not preserve enough information to answer such questions. On a development benchmark built from a real communication network, TGMS with a 14B open-source model reaches 0.409 exact match. Vector-RAG, static-graph RAG, and text-to-Cypher reach 0.045--0.182 under the same serving setup. TGMS reaches 0.67 exact match on correction probes, while the three 14B baselines score zero. The claim verifier detects all 500 injected count and entity errors with no false positives on the clean answers. Two implementation findings were especially important. First, operator output contracts prevent plans from referring to fields that do not exist. Second, verification must track whether the cited evidence is complete, because correct arithmetic over a truncated result is still misleading. The code, benchmark, and trace viewer are open source under Apache-2.0.
arXiv:2607.10412v1 Announce Type: new
Abstract: Reliable correlations of Charpy impact test results between sub-sized and full-sized specimens are essential for structural integrity assessments, particularly in nuclear applications, where spatial constraints and limited material volume restrict specimen size. Although standards such as ASTM A370 and BS 7910 provide guidance on conversion methodologies, and numerous analytical correlation methods have been proposed in prior studies, these approaches generally have limited accuracy and their applicability is often constrained to specific materials, treatment conditions, and specimen geometries. In this study, a Machine Learning (ML)-based framework is proposed for correlating Charpy impact properties across specimen sizes. The proposed approach maps absorbed energy values across the full ductile-to-brittle transition region by applying a temperature shift combined with scaled residual projection, to align sub-sized test data with full-sized response. From the resulting temperature-energy profiles, the correlated values for upper shelf energy (USE) and ductile-to-brittle transition temperature (DBTT) are extracted by fitting data with a hyperbolic tangent model. The framework is validated using a dataset comprising 389 matched sub-sized and full-sized Charpy impact tests on SA533B steel. This ML-based approach demonstrates an improved correlation performance relative to conventional analytical methods, achieving R2 values of 0.942 for USE and 0.892 for DBTT. The trained ML models do not require access to full-sized Charpy data during inference, making this approach suitable for material surveillance programs, accelerated irradiation testing, and other applications involving small-size Charpy impact testing.
arXiv:2607.10126v1 Announce Type: new
Abstract: Containerization has become increasingly essential in the machine learning (ML) domain, providing reproducibility, portability, and environment consistency. While prior studies have analyzed Dockerfile structures and best practices, none have examined ML projects in depth to reveal how the iterative nature of ML workflows influences container footprint, build performance, and caching behavior. We present the first large scale empirical study of 1,993 ML related Dockerfiles, combining quantitative analysis of container roles in ML projects and build dynamics with a qualitative investigation of refactoring practices. Results show that containers serve distinct roles across training, inference, and infrastructure. Containers are typically large, averaging 10.27 GB in size, and require long build times of about 8.84 minutes. We find that 44.4% of commits trigger rebuilds, primarily due to context file changes (96.4%), with experimentation being the main motive behind those commits that initiate rebuilds. Despite partial cache reuse, 71% of rebuild work is wasted on redundant computation. From stable projects, we identify 7 recurring ML-specific Dockerfile refactoring patterns that improve build efficiency and reduce container footprint.