CyberSec Research

Financial disclosures such as 10-K filings present challenging retrieval problems due to their length, regulatory section hierarchy, and domain-specific language, which standard retrieval-augmented generation (RAG) models underuse. We introduce FinGEAR (Financial Mapping-Guided Enhanced Answer Retrieval), a retrieval framework tailored to financial documents. FinGEAR combines a finance lexicon for Item-level guidance (FLAM), dual hierarchical indices for within-Item search (Summary Tree and Question Tree), and a two-stage cross-encoder reranker. This design aligns retrieval with disclosure structure and terminology, enabling fine-grained, query-aware context selection. Evaluated on full 10-Ks with queries aligned to the FinQA dataset, FinGEAR delivers consistent gains in precision, recall, F1, and relevancy, improving F1 by up to 56.7% over flat RAG, 12.5% over graph-based RAGs, and 217.6% over prior tree-based systems, while also increasing downstream answer accuracy with a fixed reader. By jointly modeling section hierarchy and domain lexicon signals, FinGEAR improves retrieval fidelity and provides a practical foundation for high-stakes financial analysis.

In plasma edge simulations, kinetic Monte Carlo (MC) is often used to simulate neutral particles and estimate source terms. For large-sized reactors, like ITER and DEMO, high particle collision rates lead to a substantial computational cost for such schemes. To address this challenge, an asymptotic-preserving kinetic-diffusion Monte Carlo (KDMC) simulation method and a corresponding fluid estimation technique have been proposed in the literature. In this work, we perform numerical analysis on the convergence of KDMC with the fluid estimation. To do so, we compare the accuracy of the analyzed algorithm with the accuracy of an approximate fluid method using the kinetic MC method as a reference. In a one-dimensional test case, KDMC with the fluid estimation achieves at least one order of magnitude lower errors than the fluid method for both high- and low-collisional regimes. Moreover, KDMC with the fluid estimation outperforms the kinetic MC method with a clear speed-up. Overall, our analysis confirms the effectiveness of the discussed algorithm.

We propose a computational framework, Hetero-EUCLID, for segmentation and parameter identification to characterize the full hyperelastic behavior of all constituents of a heterogeneous material. In this work, we leverage the Bayesian-EUCLID (Efficient Unsupervised Constitutive Law Identification and Discovery) framework to efficiently solve the heterogenized formulation through parsimonious model selection using sparsity-promoting priors and Monte Carlo Markov Chain sampling. We utilize experimentally observable 3D surface displacement and boundary-averaged force data generated from Finite Element simulations of non-equi-biaxial tension tests on heterogeneous specimens. The framework broadly consists of two steps -- residual force-based segmentation, and constitutive parameter identification. We validate and demonstrate the ability of the proposed framework to segment the domain, and characterize the constituent materials on various types of thin square heterogeneous domains. We validate of the framework's ability to segment and characterize materials with various levels of displacement noises and non-native mesh discretizations, i.e, using different meshes for the forward FE simulations and the inverse EUCLID problem. This demonstrates Hetero-EUCLID framework's applicability in Digital Image/Volume Correlation-based experimental scenarios. Furthermore, the proposed framework performs successful segmentation and material characterizations based on data from a single experiment, thereby making it viable for rapid, interpretable model discovery in domains such as aerospace and defense composites and for characterization of selective tissue stiffening in medical conditions such as fibroatheroma, atherosclerosis, or cancer.

Very-low-field MRIs are becoming increasingly popular due to their portability and adaptability to different environments. They are being successfully used for various clinical applications, leading to a paradigm shift in the way imaging care is typically performed. The development of low-cost MRI scanner prototypes began a few years ago, with some interesting and promising open-source projects emerging in both hardware and software design. Using permanent magnets (PMs) to generate the static magnetic field B0 can substantially reduce the manufacturing cost of low-field scanners while achieving satisfactory homogeneity. This article focuses on characterizing magnet performance in terms of B0 spatial homogeneity. Specifically, it investigates its sensitivity to various factors and explores the reasons for discrepancies between numerical expectations and actual measurements on fabricated magnets. The analysis also examines the consequences of using different numerical model approximations, revisiting concepts most frequently used in other design contexts. While these assumptions simplify the numerical model and may improve its performance in terms of computational time, this paper demonstrates that they also impact the reliability of the obtained results.

This paper proposes a human-centered conceptual model integrating lean and Industry 4.0 based on the literature review and validated it through a case study in the context of an advanced automotive first-tier supplier. Addressing a significant gap in existing research on lean Industry 4.0 implementations, the study provides both theoretical insights and practical findings. It emphasizes the importance of a human-centered approach, identifies key enablers and barriers. In the implementation process of the case study, it is considered at group level and model site level through operational, social and technological perspectives in a five-phase multi-method approach. It shows what effective human-centered lean Industry 4.0 implementation look like and how advanced lean tools can be digitized. It highlights 26 positive and 10 negative aspects of the case and their causal relation. With the appropriate internal and external technological knowhow and people skills, it shows how successful implementation can benefit the organization and employees based on the conceptual model that serves as a first step toward lean Industry 5.0.

Anti-money laundering (AML) research is constrained by the lack of publicly shareable, regulation-aligned transaction datasets. We present AMLNet, a knowledge-based multi-agent framework with two coordinated units: a regulation-aware transaction generator and an ensemble detection pipeline. The generator produces 1,090,173 synthetic transactions (approximately 0.16\% laundering-positive) spanning core laundering phases (placement, layering, integration) and advanced typologies (e.g., structuring, adaptive threshold behavior). Regulatory alignment reaches 75\% based on AUSTRAC rule coverage (Section 4.2), while a composite technical fidelity score of 0.75 summarizes temporal, structural, and behavioral realism components (Section 4.4). The detection ensemble achieves F1 0.90 (precision 0.84, recall 0.97) on the internal test partitions of AMLNet and adapts to the external SynthAML dataset, indicating architectural generalizability across different synthetic generation paradigms. We provide multi-dimensional evaluation (regulatory, temporal, network, behavioral) and release the dataset (Version 1.0, https://doi.org/10.5281/zenodo.16736515), to advance reproducible and regulation-conscious AML experimentation.

We introduce a transparent, encoding-agnostic framework for determining when the Capacitated Vehicle Routing Problem (CVRP) can achieve early quantum advantage. Our analysis shows this is unlikely on noisy intermediate scale quantum (NISQ) hardware even in best case scenarios that use the most qubit-efficient direct encodings. Closed-form resource counts, combined with recent device benchmarks, yield three decisive go/no-go figures of merit: the quantum feasibility point and the qubit- and gate-feasibility lines, which place any CVRP instance on a single decision diagram. Contrasting a direct QUBO mapping with a space-efficient higher-order (HOBO) encoding reveals a large gap. Applied to early-advantage benchmarks such as Golden-5, our diagram shows that HOBO circuits require only 7,685 qubits, whereas comparable QUBO encodings still exceed 200,000 qubits. In addition to identifying candidate instances for early quantum advantage in CVRP, the framework provides a unifying go/no-go metric that ingests any CVRP encoding together with any hardware profile and highlights when quantum devices could challenge classical heuristics. Quantum advantage in CVRP would likely require innovative problem decomposition techniques.

Software development has entered a new era where large language models (LLMs) now serve as general-purpose reasoning engines, enabling natural language interaction and transformative applications across diverse domains. This paradigm is now extending into computer-aided engineering (CAE). Recent applications of LLMs in CAE have successfully automated routine tasks, including CAD model generation and FEM simulations. Nevertheless, these contributions, which primarily serve to reduce manual labor, are often insufficient for addressing the significant computational challenges posed by large-scale, high-dimensional systems. To this aim, we first introduce the concept of LLM-empowered CAE agent, where LLMs act as autonomous collaborators that plan, execute, and adapt CAE workflows. Then, we propose an LLM-empowered CAE agent for data-free model order reduction (MOR), a powerful yet underused approach for ultra-fast large-scale parametric analysis due to the intrusive nature and labor-intensive redevelopment of solvers. LLMs can alleviate this barrier by automating derivations, code restructuring, and implementation, making intrusive MOR both practical and broadly accessible. To demonstrate feasibility, we present an LLM-empowered CAE agent for solving ultra-large-scale space-parameter-time (S-P-T) physical problems using Tensor-decomposition-based A Priori Surrogates (TAPS). Our results show that natural language prompts describing parametric partial differential equations (PDEs) can be translated into efficient solver implementations, substantially reducing human effort while producing high-fidelity reduced-order models. Moreover, LLMs can synthesize novel MOR solvers for unseen cases such as nonlinear and high-dimensional parametric problems based on their internal knowledge base. This highlights the potential of LLMs to establish the foundation for next-generation CAE systems.

Developing professional, structured reasoning on par with human financial analysts and traders remains a central challenge in AI for finance, where markets demand interpretability and trust. Traditional time-series models lack explainability, while LLMs face challenges in turning natural-language analysis into disciplined, executable trades. Although reasoning LLMs have advanced in step-by-step planning and verification, their application to risk-sensitive financial decisions is underexplored. We present Trading-R1, a financially-aware model that incorporates strategic thinking and planning for comprehensive thesis composition, facts-grounded analysis, and volatility-adjusted decision making. Trading-R1 aligns reasoning with trading principles through supervised fine-tuning and reinforcement learning with a three-stage easy-to-hard curriculum. Training uses Tauric-TR1-DB, a 100k-sample corpus spanning 18 months, 14 equities, and five heterogeneous financial data sources. Evaluated on six major equities and ETFs, Trading-R1 demonstrates improved risk-adjusted returns and lower drawdowns compared to both open-source and proprietary instruction-following models as well as reasoning models. The system generates structured, evidence-based investment theses that support disciplined and interpretable trading decisions. Trading-R1 Terminal will be released at https://github.com/TauricResearch/Trading-R1.

The petroleum industry faces unprecedented challenges in reservoir management, requiring rapid integration of complex multimodal datasets for real-time decision support. This study presents a novel integrated framework combining state-of-the-art large language models (GPT-4o, Claude 4 Sonnet, Gemini 2.5 Pro) with advanced prompt engineering techniques and multimodal data fusion for comprehensive reservoir analysis. The framework implements domain-specific retrieval-augmented generation (RAG) with over 50,000 petroleum engineering documents, chain-of-thought reasoning, and few-shot learning for rapid field adaptation. Multimodal integration processes seismic interpretations, well logs, and production data through specialized AI models with vision transformers. Field validation across 15 diverse reservoir environments demonstrates exceptional performance: 94.2% reservoir characterization accuracy, 87.6% production forecasting precision, and 91.4% well placement optimization success rate. The system achieves sub-second response times while maintaining 96.2% safety reliability with no high-risk incidents during evaluation. Economic analysis reveals 62-78% cost reductions (mean 72%) relative to traditional methods with 8-month payback period. Few-shot learning reduces field adaptation time by 72%, while automated prompt optimization achieves 89% improvement in reasoning quality. The framework processed real-time data streams with 96.2% anomaly detection accuracy and reduced environmental incidents by 45%. We provide detailed experimental protocols, baseline comparisons, ablation studies, and statistical significance testing to ensure reproducibility. This research demonstrates practical integration of cutting-edge AI technologies with petroleum domain expertise for enhanced operational efficiency, safety, and economic performance.

In the area of bearing fault diagnosis, deep learning (DL) methods have been widely used recently. However, due to the high cost or privacy concerns, high-quality labeled data are scarce in real world scenarios. While few-shot learning has shown promise in addressing data scarcity, existing methods still face significant limitations in this domain. Traditional data augmentation techniques often suffer from mode collapse and generate low-quality samples that fail to capture the diversity of bearing fault patterns. Moreover, conventional convolutional neural networks (CNNs) with local receptive fields makes them inadequate for extracting global features from complex vibration signals. Additionally, existing methods fail to model the intricate relationships between limited training samples. To solve these problems, we propose an advanced data augmentation and contrastive fourier convolution framework (DAC-FCF) for bearing fault diagnosis under limited data. Firstly, a novel conditional consistent latent representation and reconstruction generative adversarial network (CCLR-GAN) is proposed to generate more diverse data. Secondly, a contrastive learning based joint optimization mechanism is utilized to better model the relations between the available training data. Finally, we propose a 1D fourier convolution neural network (1D-FCNN) to achieve a global-aware of the input data. Experiments demonstrate that DAC-FCF achieves significant improvements, outperforming baselines by up to 32\% on case western reserve university (CWRU) dataset and 10\% on a self-collected test bench. Extensive ablation experiments prove the effectiveness of the proposed components. Thus, the proposed DAC-FCF offers a promising solution for bearing fault diagnosis under limited data.

Soil liquefaction remains an important and interesting problem that has attracted the development of enumerable prediction models. Increasingly, these models are utilizing algorithmic learning, or "artificial intelligence" (AI). The rapid growth of AI in the liquefaction literature is unsurprising, given its ease of implementation and potential advantages over traditional statistical methods. However, AI liquefaction models have been widely ignored by practitioners and researchers alike; the objective of this paper is to investigate "why?" Through a sample review of 75 publications, we identify several good reasons. Namely, these models frequently: (i) are not compared to state-of-practice models, making it unclear why they should be adopted; (ii) depart from best practices in model development; (iii) use AI in ways that may not be useful; (iv) are presented in ways that overstate their complexity and make them unapproachable; and (v) are discussed but not actually provided, meaning that no one can use the models even if they wanted to. These prevailing problems must be understood, identified, and remedied, but this does not mean that AI itself is problematic, or that all prior efforts have been without merit or utility. Instead, understanding these recurrent shortcomings can help improve the direction and perceptions of this growing body of work. Towards this end, we highlight papers that are generally free from these shortcomings, and which demonstrate applications where AI is more likely to provide value in the near term: permitting new modeling approaches and potentially improving predictions of liquefaction phenomena.

Recent large-magnitude earthquakes have demonstrated the damaging consequences of soil liquefaction and reinforced the need to understand and plan for liquefaction hazards at a regional scale. In the United States, the Pacific Northwest is uniquely vulnerable to such consequences given the potential for crustal, intraslab, and subduction zone earthquakes. In this study, the liquefaction hazard is predicted geospatially at high resolution and across regional scales for 85 scenario earthquakes in the states of Washington and Oregon. This is accomplished using an emergent geospatial model that is driven by machine learning, and which predicts the probability of damaging ground deformation by surrogating state-of-practice geotechnical models. The adopted model shows improved performance and has conceptual advantages over prior regional-scale modeling approaches in that predictions (i) are informed by mechanics, (ii) employ more geospatial information using machine learning, and (iii) are geostatistically anchored to known subsurface conditions. The utility of the resulting predictions for the 85 scenarios is then demonstrated via asset and network infrastructure vulnerability assessments. The liquefaction hazard forecasts are published in a GIS-ready, public repository and are suitable for disaster simulations, evacuation route planning, network vulnerability analysis, land-use planning, insurance loss modeling, hazard communication, public investment prioritization, and other regional-scale applications.

Using machine learning (ML), high performance computing, and a large body of geospatial information, we develop surrogate models to predict soil liquefaction across regional scales. Two sets of models - one global and one specific to New Zealand - are trained by learning to mimic geotechnical models at the sites of in-situ tests. Our geospatial approach has conceptual advantages in that predictions: (i) are anchored to mechanics, which encourages more sensible response and scaling across the domains of soil, site, and loading characteristics; (ii) are driven by ML, which allows more predictive information to be used, with greater potential for it to be exploited; (iii) are geostatistically updated by subsurface data, which anchors the predictions to known conditions; and (iv) are precomputed everywhere on earth for all conceivable earthquakes, which allows the models to be executed very easily, thus encouraging user adoption and evaluation. Test applications suggest that: (i) the proposed models outperform others to a statistically significant degree; (ii) the geostatistical updating further improves performance; and (iii) the anticipated advantages of region-specific models may largely be negated by the benefits of learning from larger global datasets. These models are best suited for regional-scale liquefaction hazard simulation and near-real-time response and are accompanied by variance products that convey where, and to what degree, the ML-predicted liquefaction response is influenced by local geotechnical data.

Mesh-based graph neural networks (GNNs) have become effective surrogates for PDE simulations, yet their deep message passing incurs high cost and over-smoothing on large, long-range meshes; hierarchical GNNs shorten propagation paths but still face two key obstacles: (i) building coarse graphs that respect mesh topology, geometry, and physical discontinuities, and (ii) maintaining fine-scale accuracy without sacrificing the speed gained from coarsening. We tackle these challenges with M4GN, a three-tier, segment-centric hierarchical network. M4GN begins with a hybrid segmentation strategy that pairs a fast graph partitioner with a superpixel-style refinement guided by modal-decomposition features, producing contiguous segments of dynamically consistent nodes. These segments are encoded by a permutation-invariant aggregator, avoiding the order sensitivity and quadratic cost of aggregation approaches used in prior works. The resulting information bridges a micro-level GNN, which captures local dynamics, and a macro-level transformer that reasons efficiently across segments, achieving a principled balance between accuracy and efficiency. Evaluated on multiple representative benchmark datasets, M4GN improves prediction accuracy by up to 56% while achieving up to 22% faster inference than state-of-the-art baselines.

Cerebral autoregulation (CA) is a fundamental mechanism that modulates cerebrovascular resistance, primarily by regulating the diameter of small cerebral vessels to maintain stable cerebral blood flow (CBF) in response to fluctuations in systemic arterial pressure. However, the clinical understanding of CA remains limited due to the intricate structure of the cerebral vasculature and the challenges in accurately quantifying the hemodynamic and physiological parameters that govern this autoregulatory process. Method: In this study, we introduced a novel numerical algorithm that employs three partial differential equations and one ordinary differential equation to capture both the spatial and temporal distributions of key CA-driving factors, including the arterial pressure (P) and the partial pressures of oxygen (PO_2) and carbon dioxide (PCO_2) within the cerebral vasculature, together with a Windkessel model in turn to regulate the CBF based on the calculated P, PO_2, and PCO_2. This algorithm was sequentially integrated with our previously developed personalized 0D-1D multi-dimensional model to account for the patient-specific effects. Results: The integrated framework was rigorously validated using two independent datasets, demonstrating its high reliability and accuracy in capturing the regulatory effects of CA on CBF across a range of physiological conditions. Conclusion: This work significantly advances our understanding of CA and provides a promising foundation for developing hemodynamic-based therapeutic strategies aimed at improving clinical outcomes in patients with cerebrovascular disorders.

This paper presents TubeBEND, a real-world dataset comprising 318 rotary tube bending processes, which were collected and sorted by experts from various fields to evaluate machine learning and signal analysis methods. The dataset addresses the industrial challenge of predicting the geometry of a first-stage bend, which can be beneficial for designing machine clamping molds for the second-stage bend in two-stage rotary draw bending. Some geometry criteria, such as the tube's final bent angle (or springback) and its cross-sectional deformation, are being recorded in this dataset. This dataset gives us the possibility to build and test machine learning models that can predict the geometry and help the machine operators with a better machine setup to optimize the tube's springback and deformation. Moreover, by recording some process parameters, such as tool movements and forces or torques applied to them, we deliver detailed information about their impacts on the final tube geometry. The focus of our work is to discover solutions that can replace traditional methods, such as trial-and-error or simulation-based predictions, by including experimental process variables in ML algorithms. Our dataset is publicly available at https://github.com/zeyneddinoz/tubebend and https://zenodo.org/records/16614082 as a benchmark to improve data-driven methods in this field.

This study explores the use of deep reinforcement learning (DRL) for active flow control (AFC) to reduce flow separation on wings at high angles of attack. Concretely, here the DRL agent controls the flow over the three-dimensional NACA0012 wing section at the Reynolds number Re = 1,000 and angle of attack AoA = 20 degrees, autonomously identifying optimal control actions through real-time flow data and a reward function focused on improving aerodynamic performance. The framework integrates the GPU-accelerated computational fluid dynamics (CFD) solver SOD2D with the TF-Agents DRL library via a Redis in-memory database, enabling rapid training. This work builds on previous DRL flow-control studies, demonstrating DRL potential to address complex aerodynamic challenges and push the boundaries of traditional AFC methods.