Graph invariant learning (GIL) seeks invariant relations between graphs and labels under distribution shifts. Recent works try to extract an invariant subgraph to improve out-of-distribution (OOD) generalization, yet existing approaches either lack explicit control over compactness or rely on hard top-$k$k selection that shrinks the solution space and is only partially differentiable. In this paper, we provide an in-depth analysis of the drawbacks of some existing works and propose a few general principles for invariant subgraph extraction: 1) separability, as encouraged by our sparsity-driven mechanism, to filter out the irrelevant common features; 2) softness, for a broader solution space; and 3) differentiability, for a soundly end-to-end optimization pipeline. Specifically, building on optimal transport, we propose Graph Sinkhorn Attention (GSINA), a fully differentiable, cardinality-constrained attention mechanism that assigns sparse-yet-soft edge weights via Sinkhorn iterations and induces node attention. GSINA provides explicit controls for separability and softness, and uses a Gumbel reparameterization to stabilize training. It convergence behavior is also theoretically studied. Extensive empirical experimental results on both synthetic and real-world datasets validate its superiority.

Bayesian optimization and progressive fine-tuning pipeline for kidney CT-scan image detection

Bond dissociation energy prediction for the carbon halogen (C-X) bond is quite important in chemistry, due to range applications of C-X bond in the drug design, reaction mechanism and material sciences fields. In the present research, a robust machine learning workflow was explored, to accurately predict the bond dissociation energy values of C-X bond. For the systematic identification of the optimized LightGBM Regressor as the top performing model, the automated machine leaning (Automl) framework, the Tree Based Pipeline Optimization Tool (TPOT) was employed. Additionally, tenfold cross-validation was used to rigorously confirm the model's robustness. The final model exhibited outstanding predictive capability, with a coefficient of determination (R2) of 0.93 on the internal test set, and 0.95 on a more stringent external validation set. Moreover, interpretation of the model via SHapley Additive exPlanations (SHAP) suggests that the model predictions are based on chemically intuitive concepts, including electronegativity difference, halogen atomic number, and local atomic charges. This work thus provides a tool for bond dissociation energy prediction that is both highly accurate and interpretable, while simultaneously demonstrating a powerful contemporary workflow for producing machine learning models that are interpretable for basic problems in chemistry.

Generative deep learning-driven de novo design of targeted MAP4K6 inhibitors.

Electrochemical and stress corrosion behaviors of low-alloy high-strength steel in the soil environment of Western China

Quantitative analysis of bacterial dynamics in time-lapse microscopy requires robust tracking pipelines, yet selecting and optimizing algorithms for specific experiments remains challenging. Indeed, microbiologists are confronted with numerous algorithms that must be carefully chosen and parameterized to achieve optimal tracking for their experiments. We present an automated methodology to determine optimal tracking configurations for microbiological applications. It is based on TrackMate 8, a novel version of the TrackMate Fiji plugin extended with microbiology-specific tools. Our approach systematically evaluates algorithm-parameter combinations optimizing biologically relevant metrics (e.g., cell-cycle accuracy, bacterial morphology) and includes: (1) integration of deep-learning algorithms (Omnipose, YOLO, Trackastra) adequate for bacterial images in TrackMate; (2) a TrackMate-Helper extension for parameter optimization; and (3) a tracking and segmentation editor for tracking ground-truth generation. We demonstrate the effectiveness of the methodology on two use cases showing its adaptability to diverse experimental conditions. This methodology enables microbiologists with a widely applicable, automated framework to optimize tracking pipelines, facilitating quantitative analysis in bacterial imaging. © 2026 Wiley Periodicals LLC. Basic Protocol 1: Impact of the respiratory chain in the growth and morphology of Escherichia coli Basic Protocol 2: The impact of mutations on the motility of Helicobacter pylori.

Industry 5.0 envisions a cyber-physical future where humans and robots collaborate harmoniously, empowered by 6G connectivity and intelligent automation. Central to this vision is the ability to autonomously configure complex production pipelines based on diverse and evolving human intents. Existing orchestration technologies exhibit critical shortcomings in terms of self-learning, validation, error diagnosis, and rectification capabilities. To this end, we propose an Agentic AI orchestration framework that interprets human intents and dynamically assembles optimal technology pipelines using a self-improving, retrieval-augmented Large Language Model (LLM) and a Bayesian contextual-bandit selector. This enables dynamic adaptation in unpredictable factory environments. Our solution is validated in a cyber-physical testbed integrating Digital Twins (DTs), distributed AI, robotics, and real-world network infrastructure. Compared to baseline LLMs, our system reduces orchestration iterations by over 94% for a given intent and by around 90% for an unseen intent, showing rapid convergence and strong generalization. Real-world deployments mirror DT results, confirming both the fidelity of the simulation and the practical value of intent-driven orchestration for human-centric manufacturing.

Systematic investigation of preprocessing pipeline for MALDI data

Autonomous underwater vehicles (AUVs) are seeing increasingly widespread adoption in marine exploration, search and rescue, and military applications. As a core enabling technology, underwater path planning faces significant challenges, in this article an intelligent path planning algorithm named A*-MPPO-DWA is proposed to enhance the efficiency and accuracy of path planning for AUV in complex dynamic environments. The proposed hierarchical framework operates as follows: firstly, the A* algorithm performs global path search and preliminary planning to ensure a feasible route from the start to the goal point. Secondly, the MPPO (multiphase path optimization) strategy refines the path through multiphase decision making. Different from conventional path smoothing and single-stage optimization methods, MPPO integrates dynamic obstacle detection and ocean current compensation into a three-stage progressive optimization pipeline, which realizes global topology preservation, redundant node elimination, and adaptive smooth correction simultaneously, rather than simple geometric smoothing. It can effectively handle complex dynamic environment. Finally, the DWA (Dynamic Window Approach) algorithm is employed for local path smoothing and real-time obstacle avoidance by integrating adaptive velocity control, enabling the AUV to avoid collisions and excessive steering during mission execution. Experimental results demonstrate that the proposed algorithm achieves superior stability and accuracy. Against baseline approaches using A* or DWA, the A*-MPPO-DWA algorithm shows significant advantages in key metrics, including path length, number of path turns, obstacle avoidance success rate, and computational time.

Vietnam's Electric Vehicle (EV) market is expanding rapidly, yet public charging infrastructure development lags significantly, exhibiting pronounced spatial imbalance in dense urban cores. This study addresses this gap through an integrated demand forecasting and location optimization framework for District 1, Ho Chi Minh City. We develop a log-linear regression model using Vietnam's macroeconomic data (2003–2023), identifying GDP and CPI as dominant determinants of vehicle ownership (R$^2$ = 0.962). Forecasted vehicle stocks for 2026–2030 are translated into public charging demand through vehicle-type disaggregation and service-capacity modeling. Spatially, we propose a four-stage optimization pipeline: demand point generation → K-Means spatial zoning → P-Median location optimization → budget-constrained allocation. Results reveal a multi-tier station distribution centered on commercial cores (Nguyen Hue–Ben Thanh corridor) with supplementary coverage in residential and tourist zones. Core areas require fast-charging expansion to manage 12% annual demand growth, while northern/southern residential blocks need baseline coverage to eliminate service gaps. Critically, we demonstrate that profit-driven private deployment creates spatial inequity, necessitating government-coordinated planning to balance efficiency and accessibility. Despite data constraints, our framework provides a replicable methodology for charging infrastructure planning in high-density Southeast Asian urban contexts, with direct implications for Vietnam's green mobility transition.

When pumped from one production unit to another, molten cheese spread is a highly viscous liquid, which inevitably affects all hydraulic calculations for optimal pipeline characteristics. However, such calculations often miss out the effect of the viscosity on the pump efficiency. This article introduces a new method for determining the optimal pipe diameter to transport liquid media during food production. It takes into account the viscosity of the liquid and the technical characteristics of the rotary lobe pump. The research featured a lobe-pump hydraulic system for transporting highly viscous cheese spreads. The calculations involved the technical parameters of the lobe pump and pipelines, as well as electricity tariffs. To determine the annual electricity costs, we used electricity tariffs for small businesses as of June 2024 in three random regions of the Russian Federation. The parameters of different cheese spreads (55–95°C) came from scientific publications in the public domain. Among the various factors that affected the optimal pipe diameter, the greatest impact belonged to the temperature-related changes in viscosity. As the operation time of the lobe pump increased, so did the share of electrical energy costs. As a result, the optimal diameter of the pipeline increased significantly to compensate for the hydraulic pressure losses and energy costs. The optimal diameter also depended on the investment parameters. Bigger Life-Cycle values correlated with larger optimal pipe diameters, i. e., the reduced costs went down. Higher interest rates, on the other hand, correlated with smaller optimal pipe diameters, i. e., the reduced costs went up. In general, the overall efficiency of the pumping station depended quite strongly on all the factors featured in this research. The new method made it possible to determine the optimal pipe diameter for inter-operational transportation of cheese spreads in particular and highly viscous laminar fluids in general. It relied on viscosity values and lobe pump specifications. Numerically, it was based on a step-by-step calculation of economic and hydraulic parameters. The method demonstrated good prospects for food pipeline design.

This paper investigates the impact of nnU-Net-based automatic segmentation on the reproducibility of radiomics features and the quality of vestibular schwannoma growth prediction. The nnU-Net model was trained on 317 T1C image pairs with masks from 155 patients from the Vestibular-Schwannoma-MC-RC2 dataset using 5-fold cross-validation, achieving a mean Dice coefficient of 0.862. ICC analysis showed that 88.8% of wavelet features maintain good or excellent agreement (ICC ≥ 0.75) between manual and automatic masks. Comparison on a subset of 96 patients with growth labels showed: the previously published pipeline (Wavelet + Voting ensemble) with manual masks achieves ROC AUC = 0.742, with automatic masks - 0.639 (−14.0%). Pipeline optimization (ICC filtering, ensemble adaptation to LR + LDA) improves the result to 0.687 (−7.4%). The results determine the cost of automating the radiomics pipeline and provide recommendations for ICC filtering for clinical implementation.

Fifth-generation advanced (5G-A) networks are expected to support ultra-dense deploy- ments, cross-domain service orchestration, and stringent quality-of-service guarantees under highly dynamic traffic and channel conditions. Conventional optimization pipelines rely on single-domain measurements and reactive heuristics, which limits their ability to capture com- plex interactions among radio access, transport load, user mobility, and application-layer de- mand. This paper presents a practical AI-driven multi-source data fusion framework for intelli- gent network optimization in 5G-A systems. The framework integrates heterogeneous telemetry from gNodeB counters, user equipment traces, edge-cloud logs, and external context signals through a temporally aligned graph-feature fusion architecture. We formulate network opti- mization as a constrained sequential decision problem and design a hybrid model that combines a spatio-temporal encoder with a policy optimization layer to jointly improve throughput, la- tency, energy efficiency, and fairness.To evaluate realism and robustness, we construct a 5G-A-oriented benchmark by combin- ing OpenRAN-style KPI streams, synthetic but statistically calibrated mobility traces, and service-level traffic profiles for enhanced mobile broadband, ultra-reliable low-latency commu- nication, and massive machine-type communication slices. Experiments are conducted on a digital twin testbed with configurable load shocks and interference bursts. Compared with rep- resentative baselines including rule-based scheduling, single-source deep reinforcement learning, and transformer-only predictors, the proposed method improves weighted network utility by 12.8%, reduces 95th percentile latency by 18.6%, and increases cell-edge user throughput by 15.2%. Ablation studies confirm that temporal synchronization, cross-source attention, and constraint-aware action projection all contribute materially to final performance.The study demonstrates that multi-source fusion is not merely a modeling preference but an operational requirement for next-generation autonomous network management. We further analyze computational complexity, deployment trade-offs, and failure modes, showing that the design can meet near-real-time control loops in edge-assisted 5G-A management stacks while maintaining stable behavior under non-stationary traffic conditions.

Disclosing microbial nitrogen metabolism in groundwater via automated machine learning-based analysis: A fluorescence information-based approach.

Comparative analysis of clearing methods for 3D imaging of the vasculature in mineralized mouse tissues.

Abstract Traditional photometric stereo techniques are constrained by stringent hardware calibration requirements, pronounced sensitivity to specular highlights and cast shadows, and an inherent difficulty in reconciling reconstruction fidelity with practical usability. These limitations significantly impede their deployment in high-precision 3D facial reconstruction. To address these challenges, this study presents an uncalibrated photometric stereo framework specifically designed for facial scenarios, integrating two key modules within a two-stage optimization pipeline. In the first stage, illumination directions are estimated using a shared-weight convolutional architecture with global max-pooling and self-attention mechanisms to extract illumination-invariant features. In the second stage, PCA-based whitening followed by Huber robust regression is employed to recover surface normals under uncertain illumination conditions. Comprehensive experiments on synthetic and real-world datasets demonstrate the effectiveness of the proposed approach. On the test set, the illumination estimation network achieves a mean angular error of 7.29° for lighting direction prediction, while intensity estimation errors are reduced by 25%–33% compared with existing baselines. For surface normal reconstruction, the proposed method achieves mean angular errors of 6.7° under the Oracle-Dir setting (using ground-truth lighting directions) and 7.6° under the Est-Dir setting (using estimated lighting directions). By eliminating the need for precise hardware calibration, the framework reduces system complexity and operational cost while improving robustness, making it well suited for applications such as 3D facial recognition and digital human modeling. Future work will explore unified end-to-end optimization strategies to further improve reconstruction accuracy and generalization capability.

This paper presents a comprehensive methodology for optimizing Linux libraries to achieve maximum performance on CPU architectures such as POWER. We present a complete optimization pipeline from compiler selection and configuration through runtime profiling and manual vectorization. The approach addresses critical performance bottlenecks through systematic application of architecture-specific compiler flags, strategic dependency management, and targeted code-level optimizations. We demonstrate that proper compiler selection, leveraging optimized dependencies such as OpenBLAS, and applying manual vectorization techniques can yield performance improvements of 10-20x over baseline implementations. The methodology is validated through practical examples including matrix multiplication libraries, demonstrating measurable improvements in FLOPS and overall throughput. These findings provide actionable guidance for developers seeking to maximize CPU utilization in performance-critical Linux applications.

The article presents an approach to phishing attack detection based on the constructionand automatic optimization of machine learning pipelines using a specialized library (PhishAutoML). The urgency of the problem is due to the evolution of phishing attacks that use socialengineering methods and lexical tricks, which makes traditional static protection methods ineffective. The theoretical foundations of text vectorization using the FastText model and its application within an AutoML approach based on Bayesian optimization, which allows for the automatic selection of hyperparameters for the entire pipeline, are described. The proposedPhishAutoML concept is used to build models capable of detecting phishing based on semanticanalysis and flexibly configuring the trade-off between quality and performance. The results ofcomputational experiments are presented: final metrics of quality and performance, as well as acomparative analysis with classical (TF-IDF) and modern (DistilBERT) approaches. The conclusions confirm the effectiveness of the proposed solution (achieving a phishing detection recall of 95%, which is several times higher than alternative methods) and outline directions forits further development. However, integrating AI into traditional intrusion detection and prevention systems poses significant risks and challenges. This article explores the key technical, organizational, and ethical barriers that hinder the widespread adoption of AI-powered solutions andsuggests potential solutions to overcome them.

Elevated brain iron levels are common in Alzheimer's disease (AD). Quantitative susceptibility mapping (QSM) is an advanced MRI technique for assessing iron accumulation. The optimized QSM at 7 Tesla (7T) MRI may further improve the sensitivity to detect subtle susceptibility changes in AD. We optimized a QSM processing pipeline for 7T MRI by systematically comparing multiple reconstruction algorithms. Evaluation criteria included image quality, artifact suppression, and anatomical clarity. The finalized pipeline was applied to individuals with AD and healthy controls (HCs). The results revealed significantly elevated magnetic susceptibility values in the globus pallidus and dentate nucleus of the AD group compared to HCs. These findings were confirmed through both visual inspection and quantitative analysis of high-resolution QSM maps. This study provides a systematic evaluation and optimization of QSM processing pipelines at 7T MRI, offering improved sensitivity and reliability for detecting ad-related susceptibility alterations. Our results highlight the importance of optimizing QSM pipelines at 7T for accurate susceptibility quantification. AD patients exhibited higher susceptibility than controls in the globus pallidus (98.7-102.9 ppb) and dentate nucleus (51.1-52.8 ppb), consistent across all QSM pipelines. We identified an optimal pipeline suitable for future applications in patients with AD and other neurological conditions.

The manufacturing industrial internet (MII) is transforming traditional factories into data-driven environments. However, the dynamically varying contexts of the MII, caused by adjustment of process parameters, equipment degradation, and customized specifications, challenge the deployed machine learning models used in process modeling, variation analysis, and anomaly detection. To address this, we propose a novel approach for robust machine learning pipeline selection and adaptation in varying industrial contexts. We introduce a weighted ensemble mechanism based on Bayesian latent space model recommender systems, optimizing sparse ensemble weights across pipelines while incorporating uncertainty quantification. This enables data-driven decision making under uncertainty by automatically selecting and adapting optimal pipelines, reducing manual intervention and improving computational efficiency. We validate our methodology using real-world data from two manufacturing processes (fused deposition modeling and aerosol jet printing) and one chemometric data set (from Tecator). Results demonstrate that our approach achieves superior and more robust performance across data sets compared with traditional single-pipeline recommenders, highlighting the importance of uncertainty quantification in improving pipeline selection accuracy and robustness. History: Bianca Maria Colosimo served as the senior editor for this article. Funding: This work was supported by the National Science Foundation [Awards DMS-2413701, DMS-2124535, CMMI-2331985, and Grant CMMI-2430998] and the American Heart Association Collaborative Science [Award 23CSA1052735]. Data Ethics & Reproducibility Note: The code capsule is available on Code Ocean at https://codeocean.com/capsule/4351450/tree/v1 and in the e-Companion to this article (available at https://doi.org/10.1287/ijds.2024.0052 ).