arXiv:2605.17669v1 Announce Type: new
Abstract: The preservation and interpretation of cultural heritage increasingly rely on digital technologies, among which Knowledge Graphs (KGs) stand out for their ability to structure vast amounts of data. However, the construction and expansion of these KGs often face challenges due to the diverse and complex nature of cultural heritage information. In this paper, we propose a novel approach for extending KG resources in the domain of cultural heritage, which we applied to French data. First, we introduce a new knowledge graph in the domain of French cultural heritage, WJoconde, which is distinguished by its multimodality as it integrates both textual and image information of the entities. We further introduce three variants of WJoconde to facilitate downstream research, such as Knowledge Graph Completion (KGC). We also built a comprehensive benchmark for KGC methods on our dataset. Second, we propose a new framework for extending cultural heritage KGs using multi-modal approaches leveraging Large Language Models (LLMs) and Vision-Language Models (VLMs), which includes automated data extraction from unstructured resources combined with a special validation pipeline for grounding the output of both models, to further extend WJoconde. Our results show that by integrating the rich text and image information in cultural heritage data, we can efficiently enhance KGs with high reliability. We open-source all code and benchmark datasets with text and images, as well as the original data with an interactive access point
Science Journals
arXiv:2605.18078v1 Announce Type: new
Abstract: Multi-agent policy-gradient methods have been shown to converge locally near stable Nash equilibria. Local convergence, however, does not determine which equilibrium is reached. We study this question through basin-entry probability with respect to a target set of equilibria selected by an external criterion, such as payoff dominance. For finite-unroll Meta-MAPG, we show that the update decomposes into ordinary policy gradient plus own-learning and peer-learning corrections, with controlled sampling noise and finite-unroll bias. We identify the peer-learning correction as the main equilibrium-selection mechanism: under a local alignment condition, the probability of entering the certified attraction region of the target stable-Nash set increases, relative to ordinary policy gradient. Because persistent correction may shift zero-update points of the original game, annealing the correction after entering the basin recovers ordinary policy-gradient dynamics and inherits local stable-Nash convergence guarantees. Experiments in Stag Hunt, iterated Prisoner's Dilemma, and preliminary neural-policy coordination environments support this basin-entry view, showing increased entry into cooperative basins under peer-aware updates.
Multi-task learning on partially labeled datasets via invariant/equivariant semi-supervised learning
arXiv:2605.17624v1 Announce Type: new
Abstract: We investigate the potential of invariant and equivariant semi-supervised learning for addressing the challenges of training multi-task models on partially labeled datasets with differently structured output tasks. Specifically, we use the popular FixMatch method for invariant semi-supervised learning and its equivariant extension Dense FixMatch. We evaluate their performance on the Cityscapes and BDD100K datasets in the context of the prevalent object detection and semantic segmentation tasks in computer vision. We consider varying sizes of the subsets annotated for each task and different overlaps among them. Our results for both invariant and equivariant semi-supervised learning outperform supervised baselines in most situations, with the most significant improvements observed when fewer labeled samples are available for a task and generally better results for the latter approach. Our study suggests that invariant/equivariant learning is a promising general direction for multi-task learning from limited labeled data.
arXiv:2605.16356v1 Announce Type: cross
Abstract: Exhaled breath condensate (EBC) contains volatile metabolites and is promising for non-invasive disease diagnosis, but after decades of research spanning over 100 biomarkers and 10 diseases, no EBC-based test has reached clinical use. The measurement variability that can span orders of magnitude, far exceeding the clinically required 10%, has long been attributed to biological factors. Here, we reveal a fundamentally different origin: the collected microdroplets themselves fail to retain volatile biomarkers. By isolating volatile co-condensation and transient evaporation from biological interference, we show that EBC microdroplets smaller than 100 {\mu}m lose clinically significant volatile content within a single breath cycle. This challenges the implicit assumption underlying decades of EBC research, that condensate faithfully reflects airway lining fluid. We develop and validate a physics-based model that predicts this loss across disease-relevant biomarkers and establishes the conditions for reliable EBC sampling. This work reframes EBC variability as a solvable engineering problem rather than an inherent biological limitation.
arXiv:2605.17572v1 Announce Type: new
Abstract: Motivated by the controller placement problems in software-defined networks and the fair division principles of classical "cake cutting", we investigate the following two-player zero-sum game. In our model, a defender places a limited number of controllers on graph vertices, while an attacker deletes a limited number of vertices. The defender score is the total number of surviving vertices reachable from any remaining controller. We formalize the computational problems associated with various game dynamics (defender plays first; attacker plays first; players play simultaneously; pure or mixed strategies).
We show that these natural problems are $\mathsf{NP}$-complete or $\Sigma^\mathsf{P}_2$-complete, depending on the specific variant. These hardness results provide limitations for optimal controller placement algorithms under different notions of quality of a solution. Finally, we present structural insights that yield efficient algorithms for restricted graph classes (namely interval graphs and graphs of bounded treewidth).
arXiv:2602.17159v5 Announce Type: replace
Abstract: The paper presents a geometric duality between the spherical squared-Hellinger distance and a hyperbolic isometric invariant of the Poincare disc under the action of the general Mobius group. Motivated by the geometric connection, we propose the usage of the L2-embedded hyperbolic isometric invariant as an alternative way to quantify divergence between Gaussian measures as a contribution to information theory.
arXiv:2605.17773v1 Announce Type: new
Abstract: Accurate estimation of plant skeletal structures (e.g., branching structures) from images is essential for smart agriculture and plant science. Unlike human skeletons with fixed topology, plant skeleton estimation presents a unique challenge, i.e., estimating arbitrary tree graphs from images. To address this problem, we introduce PlantPose, a universal plant skeleton estimator via tree-constrained graph generation. PlantPose combines learning-based graph generation with traditional graph algorithms to enforce tree constraints during the training loop. To enhance the model's generalization capability, we curate a large and diverse dataset comprising real-world and synthetic plant images, along with simplified representations (e.g., sketches and abstract drawings). This dataset enables the generalized model to adapt to diverse input styles and categories of plant images while preserving topological consistency. Our approach demonstrates robust and accurate plant skeleton estimation across multiple domains, including previously unseen out-of-domain scenarios. Further analyses highlight the method's strengths and limitations in handling complex, heterogeneous data distributions. All implementations and datasets are available at https://github.com/huntorochi/PlantPose/.
arXiv:2605.17556v1 Announce Type: new
Abstract: Clay sculpting is a nuanced, artistic task involving dexterous manipulation with long-horizon planning to achieve high-level goals. As a robotics problem, we formulate clay sculpting as a shape-to-shape matching challenge. Prior deformable object manipulation work either requires retraining a policy per goal or relies on dynamics models which represent state as sparse point clouds which do not capture important clay features, such as textures, well. We present a method for modeling the dynamics of deformable materials and planning for robotic sculpting in a representation that is visually-aligned, capturing lighting and texture features. With three different deformable materials and various end-effectors, we demonstrate that our dynamics model is comparable in performance to the state-of-the-art with the added benefit of being compatible with visual planning. Our actions are represented as parametrized pushes into clay with a single end-effector, which proved to be suitable for long-horizon (>100 actions) clay relief sculptures. Lastly, we show the benefits of planning in a visually-aligned representation, but also provide analysis providing evidence as to why this representation is challenging to plan in compared to 3D representations.
arXiv:2605.18323v1 Announce Type: new
Abstract: The finite-length performance of spatially-coupled low-density parity-check (SC-LDPC) codes is strongly affected by short cycle configurations and the harmful structures induced by them. This paper studies SC-LDPC code design directly at the protograph level, where the design variables are the edge-spreading assignments specified by the partition matrix. In contrast to CLLL/Moser--Tardos based constructive frameworks for QC-SC-LDPC codes, we focus on sharper nonconstructive existence and counting bounds. By encoding cycle-activation conditions as polynomial vanishing constraints over finite grids, we apply the Combinatorial Nullstellensatz to derive sufficient memory conditions for eliminating prescribed cycle-induced harmful structures. For fully connected $(\gamma,\kappa)$ base graphs, the resulting bounds explicitly characterize the memory required to destroy all $4$-cycles as well as all $4$- and $6$-cycles, and for fixed $\gamma$, they are asymptotically tight up to a constant factor compared with known lower bounds. We further apply the Alon--F\"uredi theorem to obtain lower bounds on the number of feasible edge-spreading assignments, including an explicit counting bound for assignments that eliminate all $4$-cycles and hence yield girth at least six. These results provide a refined algebraic-combinatorial characterization of the feasible design space for high-memory SC-LDPC codes, although no corresponding construction algorithm is provided.
arXiv:2605.18303v1 Announce Type: new
Abstract: World models built on recurrent state space architectures enable efficient latent imagination, yet remain physically unstructured, producing dynamics that violate conservation and dissipative principles. We introduce a unified Port-Hamiltonian framework that remedies this through three synergistic mechanisms. First, we embed implicit physical priors into recurrent transitions by modeling projected latent evolution as action controlled energy routing governed by flow and dissipation, biasing the projected PH phase space toward a more compact and physically structured representation. Second, we develop a kinematics aware energy world model that estimates the Hamiltonian and power balance from proprioceptive observations, providing an explicit physical signal for thermodynamic reasoning. Third, leveraging these energy gradients, we establish an energy guided Actor-Critic that uses Lagrangian multipliers to regularize policy optimization toward lower energy and smoother control. Across visual control benchmarks, this paradigm not only attains superior asymptotic returns but also elevates internal simulator fidelity by establishing a tighter, lower variance alignment between imagined and real rewards, all while reducing latent phase space volume by 4.18-8.41%, energy consumption by up to 7.80%, and mean squared jerk by up to 9.38%.
arXiv:2604.07630v2 Announce Type: replace
Abstract: The final size of an earthquake typically cannot be predicted from its ongoing seismic radiation. Expanding observations reveal distinct exceptions, such as slow earthquakes, injection-induced seismicity, and earthquake swarms, in which fault slip has an upper bound. A common thread among these anomalies is the diffusive migration of their active areas. Here, we report a unified scaling relation for these diffusional earthquakes. By tracking prolonged earthquake swarms in Northeast Japan, we constrained the time evolution of their active seismicity areas and cumulative seismic moments. Their moment-duration trajectories coincide with the final states documented for global swarms and induced seismicity across various scales. When plotted as seismic moment versus seismicity area, their trajectories collapse onto those of slow earthquakes, uniformly explained by a diffusional constant-slip model. This constant-slip scaling carves out a unique class of diffusional earthquakes, where the final available seismic energy is predetermined by slip distance.
arXiv:2604.07649v4 Announce Type: replace
Abstract: Aggregating experimental data from papers enables materials scientists to build better property prediction models and to facilitate scientific discovery. Recently, interest has grown in extracting not only single material properties but also entire experimental measurements. To support this shift, we introduce LitXBench, a framework for benchmarking methods that extract experiments from literature. We also present LitXAlloy, a dense benchmark comprising 1426 total measurements from 19 alloy papers. By storing the benchmark's entries as Python objects, rather than text-based formats such as CSV or JSON, we improve auditability and enable programmatic data validation. We find that frontier language models, such as Gemini 3.1 Pro Preview, outperform existing multi-turn extraction pipelines by up to 0.37 F1. Our results suggest that this performance gap arises because extraction pipelines associate measurements with compositions rather than the processing steps that define a material.
Fine-tuning vs. In-context Learning in Large Language Models: A Formal Language Learning Perspective
arXiv:2604.23267v2 Announce Type: replace
Abstract: Large language models (LLMs) operate in two fundamental learning modes - fine-tuning (FT) and in-context learning (ICL) - raising key questions about which mode yields greater language proficiency and whether they differ in their inductive biases. Prior studies comparing FT and ICL have yielded mixed and inconclusive results due to inconsistent experimental setups. To enable a rigorous comparison, we propose a formal language learning task - offering precise language boundaries, controlled string sampling, and no data contamination - and introduce a discriminative test for language proficiency, where an LLM succeeds if it assigns higher generation probability to in-language strings than to out-of-language strings.
Empirically, we find that: (a) FT has greater language proficiency than ICL on in-distribution generalization, but both perform equally well on out-of-distribution generalization. (b) Their inductive biases, measured by the correlation in string generation probabilities, are similar when both modes partially learn the language but diverge at higher proficiency levels. (c) Unlike FT, ICL performance differs substantially across models of varying sizes and families and is sensitive to the token vocabulary of the language. Thus, our work demonstrates the promise of formal languages as a controlled testbed for evaluating LLMs, behaviors that are difficult to isolate in natural language datasets. Our source code is available at https://github.com/bishwamittra/formallm.
arXiv:2605.17923v1 Announce Type: new
Abstract: In video generation models, particularly world models, training large-scale video diffusion Transformers (such as DiT and MMDiT) poses significant computational challenges due to the extreme variance in sequence lengths within mixed-mode datasets. Existing bucket-based data loading strategies typically rely on "equal token length" constraints. This approach fails to account for the quadratic complexity of self-attention mechanisms, leading to severe load imbalance and underutilization of GPU resources.
This paper proposes \textit{AdaptiveLoad}, an integrated optimization framework consisting of two core components: (1) A dual-constraint adaptive load balancing system, which eliminates long-sequence bottlenecks by simultaneously limiting memory consumption and computational load ($B \times S^p \le M_{\text{comp}}$); (2) A fused LayerNorm-Modulate CUDA kernel, which utilizes a D-tile coalesced reduction strategy to increase throughput and alleviate memory pressure. Experimental results on the Wan 2.1 world model demonstrate that our method reduces the computational imbalance rate from 39\% to 18.9\%, improves peak VRAM utilization efficiency by 22.7\%, and achieves an overall training throughput increase of 27.2\%.
arXiv:2605.17926v1 Announce Type: new
Abstract: Large language models (LLMs) are increasingly used to generate requirements specifications, design documents, code, and test cases. In contrast, much less attention has been given to a more difficult assurance problem: statically verifying whether implemented code satisfies requirements written in natural language. Conventional static analysis tools are effective at detecting coding defects and known vulnerability patterns, but they cannot determine whether program behavior matches intended business logic. Detecting such defects requires reasoning over the specification rather than the code alone. Software testing can expose some of these mismatches, but its effectiveness depends heavily on test design, executable artifacts, and runtime environments. This article presents a two-stage LLM-based workflow for addressing this challenge in an intelligent-vehicle cybersecurity case study. In the first stage, an AI-based rule miner extracts verifiable rules from natural-language requirements while explicitly identifying ambiguity, self-contradiction, and other non-verifiable statements. In the second stage, an AI-based code auditor checks implementation evidence against the extracted rules. Instead of asking a single LLM to directly verify code against lengthy natural-language specifications, the workflow introduces a structured intermediate representation to reduce hallucination, output variability, limited explainability, and context loss. The resulting approach is a requirement-aware and semantics-aware form of static analysis that complements software testing. By analyzing requirements and source code without requiring compilation, execution, or runtime environments, the method shifts verification and validation activities left in the development lifecycle. This LLM-based static analysis is also a new approach to addressing the test oracle problem.
arXiv:2605.18292v1 Announce Type: new
Abstract: This paper presents a method that learns a regionally stable recurrent neural network model from a set of input-output data generated by an unknown dynamical system. Relying on generalized sector conditions on the deadzone activation function, we first derive sufficient conditions that guarantee forward invariance on a compact set of the state space for any inputs from a given set. Such regional properties lead to less conservative conditions compared to variants that offer a global form of stability, and are in line with the system data that is only observed regionally. Our learning method derives conditions for regional stability using a barrier function approach, leading to models equipped with a certificate of regional stability in a subset of the state space and for a given input set. We illustrate our theoretical result with a numerical example and compare it to methods that impose a global form of stability, which fail to identify the system, and with a method that imposes no stability constraints at all, which does not guarantee a stable behavior within any state or input set.
arXiv:2510.04736v2 Announce Type: replace-cross
Abstract: Conditional Value-at-Risk (CVaR) is a leading tail-risk measure in finance, central to both regulatory and portfolio optimization frameworks. Classical estimation of CVaR and its gradients relies on Monte Carlo simulation, incurring $O(1/\epsilon^2)$ sample complexity to achieve $\epsilon$-accuracy. In this work, we design and analyze a quantum subgradient oracle for CVaR minimization based on amplitude estimation. Via a tripartite proposition, we show that CVaR subgradients can be estimated with $O(1/\epsilon)$ quantum queries, even when the Value-at-Risk (VaR) threshold itself must be estimated. We further quantify the propagation of estimation error from the VaR stage to CVaR gradients and derive convergence rates of stochastic projected subgradient descent using this oracle. Our analysis establishes a near-quadratic improvement in query complexity over classical Monte Carlo. Numerical experiments with simulated quantum circuits confirm the theoretical rates and illustrate robustness to threshold estimation noise. This constitutes the first rigorous complexity analysis of quantum subgradient methods for tail-risk minimization.
arXiv:2605.18015v1 Announce Type: new
Abstract: Production log analytics in self-hosted, resource-constrained environments requires natural-language access to massive log streams without the cost of routing every query through a large language model. We present LogRouter, an end-to-end log question-answering system deployed on TUBITAK BILGEM's national big data platform that combines a PySpark-based Drain3 ingestion pipeline, GPU-accelerated embeddings, and dual-index storage in Apache Druid and PostgreSQL with pgvector. A two-level cost-aware router dispatches each query along one of four execution paths: direct response, Druid keyword search, template lookup with SQL generation, and pgvector semantic retrieval, while a Level-2 router selects either a 14B-class or 32B-class generator for the semantic path. A dedicated coder LLM handles text-to-SQL generation. We evaluate the system on four LogHub datasets (Linux, Apache, Windows, and Mac; 70 questions in total) under both an online full-pipeline configuration and an offline configuration that isolates the generator. The router reaches 88.4% mean accuracy across datasets and 94.7% on Linux, while the full pipeline attains a mean ROUGE-1 of 0.373, BERTScore of 0.879, RAGAS Faithfulness of 0.779, and an end-to-end latency of 18.6 s. In an apples-to-apples offline comparison, the routed system reduces mean latency by 55% versus a Fixed-32B baseline (46.3 s vs. 102.1 s) while preserving Answer Correctness within 5.8 points and exceeding a Fixed-14B baseline on RAGAS Faithfulness across every dataset. Cost-aware dispatching is therefore a practical mechanism for production log QA: routing recovers most of the quality of an always-32B configuration at less than half the latency, and the L1 keyword vocabulary makes that routing decision with high precision without a learned classifier.
arXiv:2605.17527v1 Announce Type: new
Abstract: Street-view imagery (SVI) is widely used to quantify key indicators of urban environment, such as green- ery, sky, or road view indices. However, existing studies largely focus on measuring current streetscapes and rarely support the generation of alternative and non-existing urban scenarios, which is a core task in geospatial disciplines such as urban planning and design. To address this gap, we propose a gener- ative multimodal AI framework that synthesizes alternative streetscapes conditioned on targeted visual metrics, enabling direct visual exploration of urban scenarios. We first construct a multimodal dataset that aligns SVIs with textual descriptions, segmentation maps, road masks, and quantitative metrics of visual elements in Chicago and Orlando. Using this dataset, we demonstrate that diffusion models can produce realistic and semantically consistent streetscape imagery while responding to both textual and imagery controls. Our quantitative evaluations show that incorporating visual controls can improve semantic consistency, reducing the LPIPS index by approximately 6% while maintaining global visual realism. In addition, overall semantic consistency increases by 23.7% in Orlando and 46.4% in Chicago, as measured by the mIoU index, with class-wise gains exceeding even 100% improvement for building view indices. Streetscape generation can be controlled in a fine-grained manner by both visual and textual prompts, and when textual and visual controls conflict, imagery controls consistently dominate, indicating a clear control hierarchy and the importance of further developing visual controls for urban scene generation. Overall, this work establishes an important benchmark for streetscape generation us- ing SVIs and diffusion models, and illustrates how generative AI can serve as a practical, scalable, and controllable approach for urban scenario exploration.
arXiv:2605.18014v1 Announce Type: new
Abstract: Spectroscopy requires high-precision wavelength discrimination but typically requires bulky, alignment-sensitive instrumentation. To address this, we present a compact computational spectrometer built from a single germanium PN photodiode. The system exploits optoelectronic chromatic dispersion (OED), a phenomenon whereby wavelength-dependent absorption depth produces carrier diffusion delays that encode spectral information as measurable RF amplitude and phase signatures in the photodiode output. We extract DC voltage, RF amplitude, and RF phase across 15 modulation frequencies (0.1-1.5 MHz), forming a 31-dimensional feature vector per optical input. Spectral reconstruction was formulated as a high-dimensional inverse problem and solved using five machine learning models, utilizing group-wavelength splitting and k-fold cross-validation to prevent spectral leakage and ensure unbiased evaluation. Across the C- and L-bands, single-wavelength reconstruction using Gaussian Process Regression (GPR) achieves an accuracy of 0.178 nm on a wavelength-grouped, held-out test set spanning seven optical power levels. Five-fold cross-validation yields a robust Root Mean Square Error (RMSE) of (0.342 +/- 0.117) nm, confirming excellent generalization under wavelength and power variations. For dual-wavelength inputs, GPR yields accuracies of 0.362 nm for the swept wavelength and 0.434 nm for the fixed wavelength. This is the first spectral reconstruction method exploiting a multi-frequency OED feature space from a single photodiode. By merging the physics of OED with data-driven learning, this work enables alignment-free, on-chip-compatible spectrometers suitable for portable optical sensing, smartphone-integrated diagnostics, and field-deployable environmental monitoring.
arXiv:2605.17508v1 Announce Type: new
Abstract: Split Federated Learning (SFL) enables privacy-preserving collaborative training by partitioning models between clients and a server. However, under non-IID data distributions, SFL often suffers from biased optimization and unstable convergence, while existing solutions largely adapt techniques from conventional federated learning. In this work, we observe that the split architecture of SFL inherently alters how client information is represented and coordinated, opening opportunities for bias compensation beyond parameter-level aggregation. Based on this insight, we propose BESplit, an architecture-aware framework that exploits the intrinsic structure of SFL to mitigate non-IID effects. First, to prevent biased local data from dominating global updates, we introduce Evidential Aggregation (EA) to perform fine-grained reweighting of client contributions based on evidential uncertainty. Second, to further reduce distributional skew, we develop Bias-Compensated Collaboration (BCC) to align split-layer representations by pairing complementary clients. Finally, Dual-Teacher Distillation (DTD) is incorporated to synchronize knowledge between decoupled client and server models, enabling independent local inference. Extensive experiments on five benchmark datasets demonstrate that BESplit consistently outperforms state-of-the-art methods in accuracy, convergence stability, and computational efficiency under diverse non-IID settings.
arXiv:2605.08794v2 Announce Type: replace
Abstract: Modern generative models can be understood as probability transport from a simple base distribution to a target data distribution. Deterministic transport models offer tractable velocity-field parameterizations, whereas stochastic generative models capture richer density evolution through drift and diffusion. Yet when stochastic dynamics are described through deterministic velocity fields, the effects of drift and diffusion are often compressed into a single effective field, obscuring the distinct roles of deterministic evolution and stochastic fluctuation. In this work, we show that the deterministic field \(b_t\) of a stochastic generative process admits a natural transport--osmotic decomposition that separates deterministic transport from stochastic, diffusion-induced effects: \(b_t = u_t + d_t\), where \(u_t\) governs marginal probability transport and \(d_t\) captures an osmotic effect induced by diffusion and determined by the marginal score. Based on this decomposition, we propose Bridge Matching, a flow-based framework for learning decomposed generative dynamics through both marginal and conditional formulations. In generative modeling experiments, we recombine the learned components as \(b_t = u_t + \lambda_d d_t\), showing that the proposed decomposition enables interpretable and controllable sampling by adjusting the osmotic contribution in probability transport.
arXiv:2605.09124v2 Announce Type: replace
Abstract: Smart contract security has progressed from vulnerability detection toward a broader research agenda that includes semantic reasoning, automated repair, adversarial robustness, and real-time exploit detection. This paper develops a capstone-oriented research narrative around four directions: foundation-model-based smart contract semantics and vulnerability reasoning [1], automated smart contract repair with formal guarantees [2], adversarial learning for robust malicious contract and transaction detection [3], and real-time transaction-level exploit detection at blockchain scale [4]. We connect these directions to two recent studies that characterize the current frontier: a diagnostic analysis of where smart contract security analyzers fall short [5] and a scalable real-time system for malicious Ethereum transaction detection [6]. The resulting framework is intended to help students formulate capstone projects that are technically grounded, empirically measurable, and aligned with contemporary smart contract security research.
arXiv:2605.17497v1 Announce Type: new
Abstract: GRPO-style RLVR trains reasoning models from multiple on-policy attempts per prompt, but typically uses these attempts only through terminal rewards. We show that a mixed group contains a richer process signal: a correct completion is a self-generated witness of how the current policy can solve the problem, while a wrong completion provides on-policy prefixes where the policy needs correction. We introduce \emph{Self-Supervised On-Policy Distillation} (SSOPD), which distills a teacher distribution conditioned on the shortest correct completion into prefixes of the longest wrong completion. This converts intra-group correct--wrong contrast into dense process supervision without external solution traces. A stopping-time view motivates the shortest-correct / longest-wrong rule as a finite-group approximation to editing persistent failures toward fast-success actions, and a prompt-level frontier weight concentrates the auxiliary loss where correct and wrong branches coexist. Across AIME 2024, AIME 2025, and HMMT 2025, SSOPD improves over GRPO in all nine model-benchmark settings. On Qwen3-8B, it reaches a macro Avg@12 of 65.6, outperforming GRPO by 1.6 points and the solution-conditioned OPSD baseline by 0.8 points. Code will be released at https://github.com/tzq1999/SSOPD.
arXiv:2605.17475v1 Announce Type: new
Abstract: Building software that is correct by construction is a long-standing goal in software engineering, as it ensures reliability during design and development rather than after deployment. Formal methods realize this vision by enabling the expression of system behavior and requirements in mathematics, thereby guaranteeing correctness through formal verification, including theorem proving and model checking. However, the steep learning curve and demand for mathematical expertise hinder the widespread adoption of formal methods. Large language models (LLMs) have recently shown promise in bridging this gap through autoformalization. However, existing LLM-based approaches are largely limited to isolated tasks, such as theorem proving without formalization or model synthesis with insufficient verification. While valuable, these efforts do not fully exploit the potential of a more comprehensive framework in which models and proofs evolve together, a process that closely reflects real-world development practice. To address this gap, we propose Event-B Agent, a novel framework inspired by the interleaved nature of software design. Given natural language requirements, Event-B Agent constructs an initial model and iteratively repairs and refines it using formal verification feedback. Refinement simplifies proof discharge, while repair of models and proofs ensures the soundness of each refinement step. Together, these two components reinforce each other to progressively improve the model quality. Evaluation across systems of varying complexity demonstrates that Event-B Agent substantially outperforms baselines in end-to-end formal model synthesis and repair, while maintaining reasonable efficiency. These results suggest that Event-B Agent is a promising step toward correct-by-construction formal model synthesis and repair.