Chen, Hang

1 files · 151 MB · zipdeclared

What's Changed Add GPT API-based multi-agent system for automated ERT workflows by @geohang with @Copilot in https://github.com/geohang/PyHydroGeophysX/pull/3 Add multi-LLM provider support and position agent system for cross-modal geophysics by @geohang with @Copilot in https://github.com/geohang/PyHydroGeophysX/pull/4 Add ClimateDataAgent for PyDaymet integration with ERT workflows by @geohang with @Copilot in https://github.com/geohang/PyHydroGeophysX/pull/5 Add ClimateDataAgent integration for cross-modal climate-ERT reasoning by @geohang with @Copilot in https://github.com/geohang/PyHydroGeophysX/pull/6 New Contributors @geohang with @Copilot made their first contribution in https://github.com/geohang/PyHydroGeophysX/pull/3 Full Changelog : https://github.com/geohang/PyHydroGeophysX/compare/1.0...v0.3.0

Tabiri, Josephine Konamah

1 files · 11 MB · zipdeclared

This repository contains photographic images from six stores used in my thesis reserach. The files document the visual material analyzed in the study.

Levine, Jonathan · MacEachern, Leonard

1 files · 510 KB · zipdeclared

Supplementary artifact for the IEEE Embedded Systems Letters manuscript "Threshold-Based Batch Normalization for Integer-Only Binarized Neural Network Inference" by Jonathan Levine and Leonard MacEachern. The package contains FPGA HLS sources, exported trained-network parameters, ECDF/statistics data, TCL reproduction scripts, selected logs and reports, and helper scripts for the MNIST, CERN jet-substructure tagging, and UNSW-NB15 workloads. It supports the paper's threshold-export formulation for integer-only dense BNN inference and the Alveo U250 HLS results reported in the manuscript. Repository: https://github.com/maceacla/tbbn-bnn-hls

Anonymus

1 files · 424 KB · zipdeclared

Analysis code for the manuscript "Bayesian sensory integration explains ball-count bias in Major League Baseball umpires" (submitted to Communications Psychology; author information withheld for double-anonymised peer review). This archive contains the complete analysis pipeline for quantifying the ball-count-dependent strike/ball decision bias of MLB home-plate umpires and explaining it with a Bayesian sensory integration model, together with the derived result files and figures reported in the manuscript. Data: MLB 2015-2024 regular and post-season games, 451,172 called pitches (four-seam fastballs thrown by right-handed pitchers to right-handed batters). Pipeline (numbered folders 00-05): - Data acquisition (Statcast pitch-tracking data and home-plate umpire assignments) - Merging and inclusion-criteria filtering - Count-wise psychometric (probit) function fits (point of subjective equality and perceptual uncertainty) - Two-component Gaussian Mixture Model of called-pitch locations (count-specific priors) - Trial-level Bayesian sensory integration model fit (count-specific vs. universal prior) - Umpire-wise individual-differences analysis (n = 89) Environment: Python 3.13 (pybaseball 2.2.7, pandas, pyarrow) and MATLAB R2022b (Statistics and Machine Learning Toolbox). Raw and intermediate pitch-tracking data are not included, owing to file size and to avoid redistributing Baseball Savant data; they can be regenerated with the scripts in steps 00-01. See the included README.md for the full folder structure, reproduction instructions, and the correspondence between output files and the values reported in the manuscript.

Bradley, Alex

1 files · 244 KB · zipdeclared

Companion code for fitting hierarchical Bayesian spatial models that calibrate sedimentary leaf-wax n-C29 alkane hydrogen isotope ratios against precipitation isotope composition and environmental covariates. Implements 14 model variants in Stan with a Matern 3/2 Gaussian process predictive-process approximation on 125 globally distributed knots. Compares spatial and non-spatial calibrations across n = 1,128 surface sediment and soil sites from 73 publications.

Madhusudan, Gujral

6 files · 29 MB · parquetdeclared

Large language models (LLMs) are trained on massive, publicly available text datasets comprising trillions of tokens, enabling them to excel at general language tasks like next-token prediction. However, LLMs often struggle with domain-specific prompts, exhibiting reduced accuracy or generating inaccurate information (hallucinations). This is because they lack sufficient subject matter expertise. Two primary approaches exist to address this limitation for augmenting LLMs knowledge: Retrieval-Augmented Generation (RAG) and fine-tuning. This presentation focuses on fine-tuning smaller LLMs with domain-specific instruct datasets using the LoRA (Low-Rank Adaptation) technique on Gaudi hardware. We will leverage publicly available LLMs and datasets from the Hugging Face Hub for this demonstration. Though it is possible to fine tune LLMs with plain text data - sourced from documents, articles, and other materials.

Krajnc, Matej · Comi, Troy · Miao, Siqi · et al.

10 files · 25 GB · csv, zipdeclared

This dataset accompanies the manuscript "A Controlled in Silico Benchmark for GNN Prediction of Tissue Dynamics." It contains model prediction outputs, trained checkpoints, train/validation/test splits, spring-embedding outputs, generated analysis figures, analysis tables, and manuscript-specific diagnostic outputs used to reproduce the post-prediction analyses and figures. The dataset is distributed as logical ZIP archives with file-level and archive-level SHA-256 checksums. For questions, contact Tomer Stern at tomers@umich.edu.

Garrett, Patrick T. · Yates III, John R.

1 files · 83 MB · zipdeclared

Added Precursor-space MS1 gates ( tdfpy.noise.gates ). Two acquisition-aware NoiseFilter s that drop MS1 signal the instrument never fragments — signal that cannot become an identification. Both convert their (m/z, 1/K0) region once to per-scan integer TOF-index intervals (via the run calibration) and test membership with a vectorised binary search; both no-op (keep everything) when the run carries no region. Ported from the dnoise Rust tool. Compose like any filter, e.g. noise=[SelectionPolygonGate(), MadThreshold(k=3)] . SelectionPolygonGate (ddaPASEF) — keeps only MS1 points inside the run's PASEF selection polygon (the "IMS PolygonFilter" read from analysis.tdf 's GroupProperties ). A generalisation of ChargeStateRegion from a single line to the real acquisition polygon. Skipped on diaPASEF (where the same property stores window quads). Padded in physical units ( mz_pad default 5 Da, im_pad default 0.05 1/K0) so an edge precursor keeps its isotopic envelope / mobility spread rather than being clipped at a hard polygon boundary; pass mz_pad=0.0, im_pad=0.0 for a hard cutoff. DiaMs1WindowGate (diaPASEF) — keeps only MS1 points inside the union of the isolation windows ( DiaFrameMsMsWindows ); everything outside is a precursor the method never isolates. Windows are padded in physical units ( mz_pad default 5 Da, im_pad default 0.05 1/K0). No-op on ddaPASEF. Both gates now no-op (keep everything) on non-MS1 frames rather than testing fragment peaks against the MS1 precursor region (which would empty an MS2 spectrum), so they are safe to leave in a noise=[...] list applied across frame types. build_window_intervals clamps a negative scan_lo to 0 and skips boxes lying wholly outside the scan range instead of letting a negative bound wrap around via Python indexing.

Fabbri, Renato

1 files · 3.5 MB · zipdeclared

BSC Lab is an open-source sensory stimulation platform with two integrated layers: a precision multi-engine audiovisual stimulation application, and a knowledge layer built on the SSTIM ontology (OWL class hierarchy, multilingual SKOS vocabulary, SHACL validation shapes, exposure module, and external alignments) published at https://w3id.org/sstim under CC BY 4.0. Each tagged release is archived here for citation.

Bradley, Alex

1 files · 12 MB · zipdeclared

R package for reconstructing precipitation hydrogen isotope ratios from sedimentary leaf-wax n-C29 alkane measurements using the spatially-aware hierarchical Bayesian calibration of Bradley (2026). Inverts the forward calibration with full uncertainty propagation, combining analytical error, residual variance, slope posterior, and the spatial Gaussian process intercept. Supports per-record change detection with autocorrelation-adjusted thresholds and a four-level claim taxonomy from measurement-level changes through uniquely- attributable precipitation-isotope claims.

Juniper L. Simonis

1 files · 86 MB · zipdeclared

Tools for interacting with the publicly available California Delta Fish Salvage Database, including continuous deployment of data access, analysis, and presentation.

Austin, Timothy · do Valle Chagas Azaneu, Marina · Roughan, Moninya

26 files · 29 MB · netcdf, pdf, pngdeclared

Data collected from a temperature mooring at Lord Howe Island maintained by UNSW Sydney and funded by Parks Australia. The mooring position is longitude = 158.97°E and latitude = -31.51°, and local depth of approximately 52 m. The data were sampled using a series of thermistors (aqualogger 520PTs) deployed on a mooring line at 4m intervals through the water column, with shallowest instrument at 13 m and deepest at 53 m. The time period spans between 14-05-2025 and 22-04-2026. IMOS standard data quality assurance and quality control processes have been followed and the data formatted following IMOS conventions. Data quality control includes automated routines and visual inspection (expert QC) and flagging of obvious errors. File are c.f. compliant NetCDF files, and file name format follows IMOS conventions and includes sampling period in the format: UNSW_Lord_Howe_Marine_Park_TZ_ yyyymmddThhmmss Z_LH050_FV01_ LH050-2511-Aqualogger-AQUAlogger-520PT16-max160m-13_END- yyyymmddThhmmssZ.

Padilla-Villanueva, Johel

1 files · 457 KB · zipdeclared

Academy Learning Tau is an open educational and research software platform that operationalizes the Systemic Tau framework and the Discrete Extramental Clock (RECD) for complex time-series analysis. It provides ordinal metrics (τ_s, nested RECD levels), classical early-warning signals for dual reading, surrogate null models, multilingual pedagogy (Spanish, English, French), and a reproducible Streamlit laboratory for teaching and exploratory research.

Aouladali, Amal · Alaoui, Souad · Hnini, Abdelhalim

4 files · 491 KB · zipdeclared

Replication code and measured artifacts for KeystrokeAuthChain: a keystroke-dynamics continuous-authentication Transformer evaluated on the public CMU Killourhy-Maxion dataset (EER 7.47%), coupled to an on-chain audit layer that anchors each authentication decision on the Ethereum Sepolia testnet for non-repudiation and third-party-verifiable auditing. Includes the recognition pipeline, paired baselines, significance tests, the audit smart contracts and gas measurements, and the real end-to-end anchoring run (10 genuine decisions, on-chain decisionHash verified 10/10). The CMU dataset is not redistributed here (see data/README.md inside the archive).

Padilla-Villanueva, Johel

1 files · 356 KB · zipdeclared

Plataforma web educativa y de investigación para el paradigma Tau Sistémica y el Reloj Extramental Discreto (RECD). Incluye un laboratorio interactivo para el análisis de señales biológicas, ecológicas y financieras mediante ordinal patterns, EWS y TDA.

Bradley, Alex

1 files · 10 MB · zipdeclared

Posterior draws from 14 hierarchical Bayesian leaf-wax-to-precipitation calibration models, fit in Stan on the frozen calibration run c2_run_20260626 (n = 1128 observations; Bradley 2026, Communications Earth and Environment). Each file is a posterior::draws_df with 1000 stratified draws across 8 MCMC chains. Spatial models include 125 per-knot latent z values on a Fibonacci sphere lattice. Used by the leafwax R package as the full-resolution backing data; the package ships a 100-draw fixture and downloads the full posteriors from this archive on first use. Supersedes the v1.x deposit (v10 / n = 1129 lineage).

Suchanek, Eric G., PhD

1 files · 5.3 MB · zipdeclared

MemoryKG builds a hybrid semantic + structural knowledge graph from Markdown and plain-text document corpora. It chunks text semantically, discovers structural and semantic relationships between sections and chunks, stores them in SQLite, and augments retrieval with vector embeddings via LanceDB. It also supports a conversational memory layer — ingesting and indexing agent turns, consolidating them into summaries, and enabling semantic recall across sessions via MCP-based AI integration.

Wang, Chang · Lu, Xingcheng

2 files · 222 MB · netcdfdeclared

Machine-learning-inferred monthly anthropogenic NOx emission over the 2026 Strait-of-Hormuz disruption (global, 0.1 degree, January 2025 - May 2026). This dataset is the top-down NOx emission product underlying the companion manuscript on the 2026 Strait-of-Hormuz shipping-emission collapse. A LightGBM estimator trained on the CAMS-GLOB-ANT v6.2 inventory (2018-2024), with the observed TROPOMI NO2 column and GEOS-CF chemistry/meteorology as predictors, is applied month by month to 2025-01 through 2026-05 to infer the anthropogenic NOx emission flux. Provided as a single self-describing CF-1.8 NetCDF containing: the total anthropogenic NOx flux (kg m-2 s-1, reported as NO) and a per-pixel cross-validation uncertainty (log-space). The estimator resolves the total emission only; no sector decomposition is distributed, because over open ocean the total is essentially ship emission while on land a sector split would only re-apply the CAMS-GLOB-ANT prior shares and is not constrained by the observations. Coverage is land and ocean within +/-60 degrees latitude on a regular 0.1-degree global grid. All units and coordinates are embedded in the file.

DHI

3 files · 28 GB · pdf, zipdeclared

MIKE 3 Flow Model FM is a 3D hydrodynamic modeling system based on a flexible mesh approach, used for oceanographic, coastal, and estuarine applications. This repository includes a model setup and 2-year model results for Øresund (the strait between Denmark and Sweden), observational data, and code for model validation. This dataset is part of the WaterBench series by DHI, supporting open research on water-related challenges. It is intended for educational and research purposes, including model validation, parameter calibration, and machine learning applications. Results should not be used for decision-making. Files: README: Description of dataset with details on citations, data processing, and background information. WaterBench-MIKE3HD-Oresund.zip : model setup, input data (e.g., boundary conditions, wind), observational data, and code for data exploration and model validation. MIKE3HD-Oresund-output.zip : 2-years model result files (~28 GB).