Assessing the maturity of data governance, data management, and data quality still relies on manual, expert-driven external assessments, which are costly and difficult to repeat over time. This paper presents DQPA, a web platform that supports diagnostic internal self-assessment in accordance with the ISO/IEC 33,000 assessment mechanism instantiated by an ISO 8000-based process reference model. A deterministic, open-source assessment engine traceably derives process capability and organizational maturity levels by applying the normative aggregation rules; a complementary service based on a pre-trained language model, conditioned by a structured prompt and grounded in an expert-curated catalogue of improvement tasks, transforms the computed state into prioritized recommendations. The engine was verified against a reference implementation in professional use and against an expert external assessment in a real organization; the generated recommendations were rated through a structured rubric by four expert evaluators. DQPA has been applied in four organizations of different sectors, ownership, and sizes, and a usability assessment based on the System Usability Scale indicated a favourable level of perceived usability.
WaveLab Studio is a lightweight Python tool for constructing, executing and visually analysing simplified two-dimensional wave-propagation experiments in heterogeneous material scenes. Its distinct contribution is an end-to-end workflow that keeps the same scene description consistent across an interactive graphical editor, a YAML serialization format, command-line execution, a headless Code Ocean run and a standardized export package containing fields, arrays, metrics and optional animations. The current release is intended for teaching, thesis supervision, rapid scene prototyping and preparation of inspectable inputs for later scientific-machine-learning experiments. It is not a full-vector Maxwell solver and does not claim physical validation against FDTD or FEM packages. This paper defines the implemented scene schema, the default scalar propagation model, the internal construction of the reference and approximated fields, the metric interpretation, the runtime context, the current status of the optional PINN/Modulus pathway and the validity limits of the material parameterization.
SnowMapPy is an open-source Python-based package developed to streamline the collection, processing, and analysis of MODIS snow cover data from Terra and Aqua satellites. By automating essential steps, including data clipping, reprojection, filtering, and time series generation, SnowMapPy significantly enhances the efficiency and precision of snow hydrology research. The package allows users to collect and process data directly from Google Earth Engine, enabling efficient data acquisition and processing tailored to the needs of snow hydrology, water resource management, and climate change studies. Designed for accessibility and flexibility, SnowMapPy supports large-scale, high-resolution snow cover analysis with minimal configuration. Through its modular structure, the package facilitates customized workflows, making it a valuable tool for researchers aiming to understand snow dynamics and its impact on seasonal water resources. The first application of SnowMapPy was in the Moroccan Atlas Mountains, where snowmelt is essential for water resources. The tool effectively captures spatial and temporal snow cover variations, offering valuable insights for water management and climate impact assessments. This initial implementation highlights its potential for broader application in other snow-covered regions worldwide.
Federated learning (FL) enables collaborative model training without requiring the exchange of raw data, making it suitable for distributed Artificial Intelligence of Things (AIoT) environments. This paper presents LIA-FL, a containerised software platform for deploying and orchestrating FL workflows using Docker, Docker Compose, MQTT-based communication, TensorFlow/Keras models, and local SQLite storage. The platform provides federated nodes, an aggregation server implementing the Federated Averaging algorithm, model persistence, fragmented model-weight transmission, integrity checking, and structured logging facilities. The software is demonstrated through a meteorological forecasting use case involving distributed weather stations organised into groups of contextually similar nodes, showing its applicability to reusable AIoT scenarios.
Since its v2.0.0 release, NEML2 has been rewritten from the ground up. Version 2 was a C++-native constitutive-modeling library deployed for fast execution through TorchScript tracing and just-in-time (JIT) compilation. Version 3 is a Python-native library whose models are authored as ordinary PyTorch modules and deployed through Ahead-of-Time Inductor (AOTI), a modern PyTorch graph-compilation backend that produces portable, Python-free artifacts loadable from pure C++. This update summarizes the rewrite and the investment that accompanied it: the AOTI export pipeline, a new pluggable solver layer with adaptive substepping, an expanded physics-model catalog, developer tooling, and the corresponding overhaul of NEML2’s MOOSE integration. A head-to-head benchmark shows AOTI outperforming legacy JIT on every GPU scenario measured, by 1.0–4.5×.
Sara is an open-source Python desktop application for creating and using interactive visual scene displays in augmentative and alternative communication. It allows speech-language therapists, educators, and researchers to build scene-based communication activities with manually configured hotspots, contextual visual supports, audio, text-to-speech output, pictographic resources, and navigation between scenes. Projects are stored as JSON files with associated media assets, so materials can be reopened, modified, shared, and reused. Sara also includes optional research logging with interaction events and professional annotations, supporting transparent documentation of how prepared visual scene display activities are used.
Robodimm is an open-source web application for programming industrial-robot motions, computing trajectory-level inverse dynamics, and screening motor–gearbox candidates from an actuator catalog. It supports a four-axis palletizer with a closed-chain parallelogram and a six-axis serial manipulator. A browser-only DEMO mode provides interactive kinematics, trajectory generation, torque logging, and preliminary sizing; a local PRO backend uses Pinocchio for higher-fidelity dynamics. Two nominal applications demonstrate cross-software verification against author-developed Simscape Multibody models: the PRO torque RMSE is 0.0408543 Nm for the palletizer and 9.00185×10−13 Nm for the serial manipulator. Robodimm reports reproducible candidate rankings, not procurement validation. The complete verification battery, diagnostic records, models, and raw results will accompany the reviewed release in a versioned reproducibility archive.
Natural language processing (NLP) systems usually analyse text by adding token, span, and dependency annotations. Professional multilingual workflows, including conference interpreting, legal translation, and cross-border M&A documentation, also require document hierarchy, version history, parallel language states, and auditable amendments. USD-NLP addresses this representation problem by adapting Universal Scene Description (USD) composition concepts to multilingual text workflows. It represents projects, documents, language or processing layers, and linguistic units as composable objects, so amendments and language-specific states can be layered without overwriting earlier versions. USD-NLP does not replace automatic annotation engines; it stores their outputs in a versioned structure. Token attributes such as lemma, part of speech, and morphology are stored as Prim metadata, named entities are represented as span Prims over token identifiers, and dependency trees are represented as typed links from dependent tokens to head tokens. A lightweight CoNLL-U adapter converts standard pipeline output into this structure while preserving the zero-dependency design of the core library. Performance benchmarks on a 30,000-Prim scene show construction throughput of approximately 92,000 Prims per second, layer composition at 1.4 ms per document stage, non-destructive override application at over 500,000 operations per second, and a memory overhead of 1.35× relative to flat dictionary baselines.
We present SDG, a Domain-Specific Language for the automated synthesis of stream data generators. SDG provides a unified framework for modeling both data and concept drifts across various scenarios, including sudden, gradual, incremental, and recurring shifts. Built using textX, the DSL transforms declarative specification models into executable Python code that generates data streams compatible with River (natively) and MOA (CSV/ARFF export). Additionally, SDG features an LLM-powered interface for converting natural language into formal specifications. By separating dataset definition from implementation, SDG ensures reproducibility and significantly streamlines the benchmarking of stream learning algorithms.
BLOCKS is an open-source, web-based application for administering blocked, text-based vignette experiments. Researchers prepare the blocked vignette design using their preferred statistical or programming software. The selected block structure and outcome measures are then implemented in BLOCKS through plain-text vignette files organised into vignette set-specific folders and structured Excel workbooks. At runtime, BLOCKS configures the online experiment from these file-based inputs, uses adaptive assignment to the least-occupied vignette set, with random tie-breaking, randomises vignette order within sets, validates responses, records attention-check results, stores data in a relational database, captures process-generated data such as response latency, click counts, answer changes, and timestamps, and provides protected CSV export. By separating researcher-defined study configuration from the web-application logic that administers the study, BLOCKS lowers implementation barriers so researchers can focus on substantive expertise. The software supports transparent and reproducible implementation through inspectable file-based inputs, version-controlled workflows, and structured data export, with open access and archival provided through GitHub and Zenodo.
Atomistic simulation workflows increasingly combine large structures with fast machine-learning interatomic potentials, making auxiliary geometric operations potential bottlenecks. AtomVoxelizer accelerates these tasks by constructing periodic voxel grids using local stencil application. Rather than evaluating every voxel-atom distance, AtomVoxelizer precomputes local sphere stencils and translates them to each atom, reducing common mask-generation scaling from O(N3) to approximately O(N) in the number of atoms at fixed cutoff radius and grid spacing by shifting the algorithm to an atom-centered picture. The package supports masks, distance fields, vector and matrix grids, connected-region and surface analysis, probe accessibility, Numba acceleration, and examples for zeolites, nanoparticles, surfaces, and voxel-guided Monte Carlo trial moves.
PyGCMS Pipeline is a browser-based workflow for nominal-mass, MS1, full-scan GC-MS chromatograms, developed and validated primarily for pyrolysis GC-MS. It reads a documented mzML subset, performs configurable chromatographic deconvolution using eleven numeric controls and a bleed-filter toggle, submits search-eligible reconstructed spectra to a locally licensed NIST MS Search installation, and preserves every returned top-N candidate with its original rank, scores and identifiers. Version 1.3.1 adds a working default example, configurable low-ion search eligibility, top-two unique-species match-gap flags, CAS-based duplicate grouping in the view, regression-tested idempotent candidate selection, and classification model 2.0.0. The final classification writer records rule, confidence, warning, lock status and model version while leaving the raw candidate list unchanged. Parameter presets, selected identities, alternatives, rationales, versions and hashes are exported together, allowing the archived candidates to be reassessed with revised retention-index evidence, classification rules or human decisions without repeating the original search. The software does not claim a new deconvolution theory or definitive compound identification; it provides an auditable implementation for complex chromatographic profiles.
EchoVerse Sonar Lab is an open-source software framework for multimodal sonar simulation, visualization, data streaming, and offline waveform/image processing. It provides a unified project workflow for forward-looking sonar, multibeam echo sounder, and side-scan sonar simulation within shared scene and sensor configurations. The interactive C++ runtime loads scenes and generates real-time geometry-derived range/intensity images and echo point clouds for display, TCP streaming, and packetized .esl3d recording. The MATLAB pipeline consumes those recorded intermediate data, synthesizes received channel echoes, exports waveform data to HDF5, and performs matched filtering, beamforming, and final waveform-domain image reconstruction. EchoVerse Sonar Lab supports reproducible sonar data generation, early-stage algorithm development, software-in-the-loop integration, and educational demonstration.
Across planetary surfaces, many landform measurements still rely on expert visual interpretation and manual mapping, introducing subjectivity and limiting both reproducibility and scalability. To address these limitations, we developed MarsCONE 2.0, an extended version of the original MarsCONE workflow designed for automated morphometric analysis of conical landforms, including Martian pitted cones and terrestrial volcanic analogues. The performance of our updated software was tested using Martian High Resolution Imaging Science Experiment (HiRISE)- and Context Camera (CTX)-derived digital elevation models (DEMs) from Ulysses Colles and Hydraotes Colles, Mars, as well as terrestrial DEM datasets from the San Francisco Volcanic Field, USA, and the Michoacán–Guanajuato Volcanic Field, Mexico. The morphometric core of this new software version builds on the previously validated MarsCONE workflow, where automatically derived cone parameters were evaluated against manual ArcGIS Pro measurements using standard agreement and error metrics. In this study, we further validated the workflow using terrestrial datasets, demonstrating its applicability beyond the original Martian implementation. MarsCONE 2.0 extends the original workflow with a PySide6-based graphical user interface (GUI), enhanced Finder and Analyzer modules, integrated quality-control tools, new Elevation Explorer and Complex Cones modules, and multi-dataset comparison capabilities. These additions extend the workflow to support the analysis of isometric cones and composite edifices comprising overlapping cones with multiple summit craters in terrestrial and planetary settings, while facilitating automated comparisons across datasets. Moreover, the GUI integrates workflow configuration, execution, visual diagnostics, plotting, and GIS-ready export while preserving the original modular processing architecture. This broadens accessibility beyond computational experts, making the workflow more approachable for Earth and planetary scientists with diverse levels of programming experience. For the illustrative example presented here, MarsCONE 2.0 produced internally consistent morphometric results across Martian and terrestrial datasets of the same regions spanning different spatial resolutions. Our updated framework therefore improves accessibility, reproducibility, and versatility of DEM-based cone morphometry while preserving the modular processing architecture of the original MarsCONE workflow. We contend that MarsCONE 2.0 provides a practical pathway from manual mapping to scalable, automated, and data-driven morphometric analysis of planetary surfaces.
The R package sglg provides a computational framework for fitting, summarizing, and evaluating parametric and semi-parametric generalized log-gamma regression models, together with several methods for their visualization. We implement inference methods based on likelihood, parametric bootstrap, and penalized likelihood frameworks. It is also possible to fit models in the presence of censored observations. The package offers the ability to fit nonparametric functions using P-splines or penalized natural cubic splines. We also implement residual tools, such as deviance and quantile residuals as well as envelope plots. Additionally, we provide a procedure for local influence analysis under two different perturbation schemes. Lastly, all the aforementioned functionalities and many other auxiliary functions are illustrated through two real data sets.
Bangla is a morphologically rich, low-resource language whose regional dialects suffer from an even greater scarcity of tools, corpora, and annotated datasets. This paper introduces brnltk, the first modular Python toolkit designed for processing Bangla regional dialects. It integrates rule-based, statistical, and neural components for dialect translation, POS tagging, stemming, tokenization, and semantic similarity computation across five regions: Mymensingh, Barishal, Chattogram, Noakhali, and Sylhet. POS tagging reaches up to 96% accuracy, exceeding a general-purpose Bangla baseline (50.8%); dialect translation attains chrF 25.0–54.1, BLEU 1.4–13.4, METEOR 11.2–37.5, WER up to 89.4%, and CER up to 64.3% across dialect pairs. Training-free for end users and built on human-verified dialectal mappings, brnltk requires no annotated corpora or fine-tuning.
In this work, we present an update of the PySensMCDA library that extends its scope from perturbation-based sensitivity analysis to inverse analysis of the weight space. Four submodules are introduced. The sampling submodule generates weight vectors under absent or partial preference information and decision matrices with uncertain criteria values. The smaa submodule implements the stochastic multicriteria acceptability analysis family, complemented by an exact linear-programming route. The eors submodule provides exhaustive objective ranking together with local sensitivity diagnostics. The robustness submodule summarises any collection of rankings through stability coefficients, decision entropy and pairwise winning indices. A shared calling convention makes every analysis usable with any MCDA method, including user-defined ones. The update also adds six visualizations, an example notebook and 371 tests.
Lidar technology has been widely used for collecting 3D urban data from long distances. Unfortunately, while high-altitude aerial scans offer an efficient method for acquiring large-scale spatial data, the obtained results are often incomplete due to limitations in sensor resolution, scanning angles and object occlusions. In recent years, the advancements in deep learning technologies led to the development of many point cloud completion methods, however these methods impose strict data size limitations for input point clouds. This paper presents a novel framework designed to overcome the input data size limitations and enhance the accuracy of existing point cloud completion models by the means of data partitioning and postprocessing. Testing has shown that the framework can generate accurate point clouds that are at least 5 times larger than the input.
This paper presents TwoTankFuzzyModel, an open MATLAB package for generating Takagi–Sugeno (TS) fuzzy models of a nonlinear two-tank benchmark commonly used in control and diagnosis research. The software implements a physically grounded nonlinear model based on mass balance and gravity-driven outflows, and constructs a TS approximation as a convex combination of locally linearized state-space models obtained via first-order Taylor expansion on a user-defined grid of operating points. The package provides reproducible workflows for: (i) parameterized model instantiation via tank and hydraulic data, (ii) automatic TS model generation and discretization, (iii) membership-function based interpolation (triangular or Gaussian), and (iv) optional illustrative LMI-based controller and observer synthesis for simulation studies. The toolbox enables rapid benchmarking and fair comparison of TS-based methods across different physical parameterizations.
