Data Pipeline Research Articles

Challenges and practices of deep learning model reengineering: A case study on computer vision

ContextMany engineering organizations are reimplementing and extending deep neural networks from the research community. We describe this process as deep learning model reengineering. Deep learning model reengineering — reusing, replicating, adapting, and enhancing state-of-the-art deep learning approaches — is challenging for reasons including under-documented reference models, changing requirements, and the cost of implementation and testing.ObjectivePrior work has characterized the challenges of deep learning model development, but as yet we know little about the deep learning model reengineering process and its common challenges. Prior work has examined DL systems from a “product” view, examining defects from projects regardless of the engineers’ purpose. Our study is focused on reengineering activities from a “process” view, and focuses on engineers specifically engaged in the reengineering process.MethodOur goal is to understand the characteristics and challenges of deep learning model reengineering. We conducted a mixed-methods case study of this phenomenon, focusing on the context of computer vision. Our results draw from two data sources: defects reported in open-source reeengineering projects, and interviews conducted with practitioners and the leaders of a reengineering team. From the defect data source, we analyzed 348 defects from 27 open-source deep learning projects. Meanwhile, our reengineering team replicated 7 deep learning models over two years; we interviewed 2 open-source contributors, 4 practitioners, and 6 reengineering team leaders to understand their experiences.ResultsOur results describe how deep learning-based computer vision techniques are reengineered, quantitatively analyze the distribution of defects in this process, and qualitatively discuss challenges and practices. We found that most defects (58%) are reported by re-users, and that reproducibility-related defects tend to be discovered during training (68% of them are). Our analysis shows that most environment defects (88%) are interface defects, and most environment defects (46%) are caused by API defects. We found that training defects have diverse symptoms and root causes. We identified four main challenges in the DL reengineering process: model operationalization, performance debugging, portability of DL operations, and customized data pipeline. Integrating our quantitative and qualitative data, we propose a novel reengineering workflow.ConclusionsOur findings inform several conclusion, including: standardizing model reengineering practices, developing validation tools to support model reengineering, automated support beyond manual model reengineering, and measuring additional unknown aspects of model reengineering.

An image processing pipeline for electron cryo-tomography in RELION-5.

Electron tomography of frozen, hydrated samples allows structure determination of macromolecular complexes that are embedded in complex environments. Provided that the target complexes may be localised in noisy, three-dimensional tomographic reconstructions, averaging images of multiple instances of these molecules can lead to structures with sufficient resolution for de novo atomic modelling. Although many research groups have contributed image processing tools for these tasks, a lack of standardisation and interoperability represents a barrier for newcomers to the field. Here, we present an image processing pipeline for electron tomography data in RELION-5, with functionality ranging from the import of unprocessed movies to the automated building of atomic models in the final maps. Our explicit definition of metadata items that describe the steps of our pipeline has been designed for interoperability with other software tools and provides a framework for further standardisation.

A Comprehensive Online Analytical System Coupled with Standardized Data Analysis for the Electrochemical Reduction of CO2

The electrochemical reduction of CO2 (CO2RR) is a promising way to convert detrimental CO2 emissions into sustainable fuels and chemicals, and thus to achieve a circular carbon economy. 1 Depending on catalyst type and reaction conditions, different gaseous and liquid products can be obtained from the CO2RR, and their production rates vary with reaction time. It becomes thus imperative, for any catalytic performance analysis, to be able to assess the product distribution online during CO2RR. 2,3 However, while gaseous products are readily analyzed online by gas chromatography, liquid products are typically only assessed at the end of the reaction due to the lack of suitable automated liquid sampling and analysis methods. In addition, reaction parameters such as CO2 mass flow rates, temperatures and pressures are seldom recorded, causing the loss of significant information on catalysts, electrodes, and electrolyzers behavior.To overcome these issues, we assemble a comprehensive analytical system coupling online gas and liquid product analysis by gas and liquid chromatography leveraging a special automated liquid sampling valve with electrochemical protocols that assess CO2RR performance, electrolyzer cell resistance and electrode surface area. In addition, we record CO2 mass flow rates, electrolyzer temperatures, as well as gas and liquid flow pressures. 4 To rapidly and reproducibly handle the large and heterogeneous data volume obtained we implement a standardized data pipeline based on our own open-source software,5 which automatically parses the numerous different raw data files, composes a data set following FAIR practices,6 and post-processes and plots the data in a standardized way. We validate the analytical system by carrying out CO2RR at 200 mA/cm2 on Cu gas diffusion electrodes, following the changes in selectivity with reaction time for > 10 gaseous and liquid products, and recording mass flow rates, electrolyzer temperatures and pressures. The modular nature of our analytical system, combined with the standardized data pipeline, allows us to freely increase the number and type of sensors used with minimal impact on the data analysis time, as well as to multiplex our analysis to 8 parallel electrolyzer cells, paving the way for a much deeper and faster understanding of the function of CO2RR catalysts, electrodes, and electrolyzers.

Common sleep data pipeline for combined data sets.

Over the past few years, sleep research has shown impressive performance of deep neural networks in the area of automatic sleep-staging. Recent studies have demonstrated the necessity of combining multiple data sets to obtain sufficiently generalizing results. However, working with large amounts of sleep data can be challenging, both from a hardware perspective and because of the different preprocessing steps necessary for distinct data sources. Here we review the possible obstacles and present an open-source pipeline for automatic data loading. Our solution includes both a standardized data store as well as a 'data serving' portion which can be used to train neural networks on the standardized data, allowing for different configuration options for different studies and machine learning designs. The pipeline, including implementation, is made public to ensure better and more reproducible sleep research.

Strategic Data Pipeline Design: Enhancing Operational Efficiency from Oracle to Single Store using Airflow S3 Data Pipelines

Strategic Data Pipeline Design: Enhancing Operational Efficiency from Oracle to Single Store using Airflow S3 Data Pipelines

Towards proactive corrosion management: A predictive modeling approach in pipeline industrial applications

Corrosion is the most significant factor affecting pipeline integrity management. It may lead to facility damage, production interruption, environmental pollution and even explosion. Corrosion management developed in recent years is a critical means to ensure the safety and reliability of pipeline systems. However, developing proactive corrosion management is difficult due to limited availability of data on target pipelines and corrosion factors. The motivation of this study is to develop a proactive corrosion management approach based on a predictive framework. The proposed framework integrates multiple corrosion factors, diverse working conditions, multi-phase degradation, and probabilistic analysis to offer an efficient analysis tool. A numerical analysis of corrosion prediction is conducted using the corrosion data of the training pipeline and the target pipeline, and the corrosion rate, corrosion depth, and remaining useful life (RUL) of the target pipeline are evaluated. Based on the RUL information, the maintenance time of the pipeline is further analyzed to maximize the benefits of maintenance. This study offers a comprehensive approach to pipeline corrosion prediction, facilitating proactive maintenance strategies and improving asset management in the pipeline industry.

Embedded FPGA developments in 130 nm and 28 nm CMOS for machine learning in particle detector readout

Embedded field programmable gate array (eFPGA) technology allows the implementation of reconfigurable logic within the design of an application-specific integrated circuit (ASIC). This approach offers the low power and efficiency of an ASIC along with the ease of FPGA configuration, particularly beneficial for the use case of machine learning in the data pipeline of next-generation collider experiments. An open-source framework called “FABulous” was used to design eFPGAs using 130 nm and 28 nm CMOS technology nodes, which were subsequently fabricated and verified through testing. The capability of an eFPGA to act as a front-end readout chip was assessed using simulation of high energy particles passing through a silicon pixel sensor. A machine learning-based classifier, designed for reduction of sensor data at the source, was synthesized and configured onto the eFPGA. A successful proof-of-concept was demonstrated through reproduction of the expected algorithm result on the eFPGA with perfect accuracy. Further development of the eFPGA technology and its application to collider detector readout is discussed.

A scalable and transparent data pipeline for AI-enabled health data ecosystems.

Transparency and traceability are essential for establishing trustworthy artificial intelligence (AI). The lack of transparency in the data preparation process is a significant obstacle in developing reliable AI systems which can lead to issues related to reproducibility, debugging AI models, bias and fairness, and compliance and regulation. We introduce a formal data preparation pipeline specification to improve upon the manual and error-prone data extraction processes used in AI and data analytics applications, with a focus on traceability. We propose a declarative language to define the extraction of AI-ready datasets from health data adhering to a common data model, particularly those conforming to HL7 Fast Healthcare Interoperability Resources (FHIR). We utilize the FHIR profiling to develop a common data model tailored to an AI use case to enable the explicit declaration of the needed information such as phenotype and AI feature definitions. In our pipeline model, we convert complex, high-dimensional electronic health records data represented with irregular time series sampling to a flat structure by defining a target population, feature groups and final datasets. Our design considers the requirements of various AI use cases from different projects which lead to implementation of many feature types exhibiting intricate temporal relations. We implement a scalable and high-performant feature repository to execute the data preparation pipeline definitions. This software not only ensures reliable, fault-tolerant distributed processing to produce AI-ready datasets and their metadata including many statistics alongside, but also serve as a pluggable component of a decision support application based on a trained AI model during online prediction to automatically prepare feature values of individual entities. We deployed and tested the proposed methodology and the implementation in three different research projects. We present the developed FHIR profiles as a common data model, feature group definitions and feature definitions within a data preparation pipeline while training an AI model for "predicting complications after cardiac surgeries". Through the implementation across various pilot use cases, it has been demonstrated that our framework possesses the necessary breadth and flexibility to define a diverse array of features, each tailored to specific temporal and contextual criteria.

Distributed edge analytics in edge‐fog‐cloud continuum

AbstractTo address the increased latency, network load and compromised privacy issues associated with the Cloud‐centric IoT applications, fog computing has emerged. Fog computing utilizes the proximal computational and storage devices, for sensor data analytics. The edge‐fog‐cloud continuum thus provides significant edge analytics capabilities for realizing interesting IoT applications. While edge analytics tasks are usually performed on a single node, distributed edge analytics proposes utilizing multiple nodes from the continuum, concurrently. This paper discusses and demonstrates distributed edge analytics from three different perspectives; serverless data pipelines (SDP), distributed computing and edge analytics, and federated learning, with our frameworks, MQTT based SDP, CANTO and FIDEL, respectively. The results produced in the paper, through different case studies, show the feasibility of performing distributed edge analytics following the three approaches, across the continuum.

Optimizing Data Pipeline Efficiency with Machine Learning Techniques

In the era of big data, efficient data processing is crucial for timely insights and decision-making. Traditional data pipelines face challenges such as latency, scalability, and fault tolerance. This paper explores the application of machine learning (ML) techniques to optimize data pipeline efficiency. We propose a framework that integrates ML models for predictive resource allocation, anomaly detection, and dynamic scaling within data pipelines. Our experiments demonstrate significant improvements in processing speed, resource utilization, and reliability. Key Words: Data Engineering, Data Pipelines, Machine Learning, Predictive Resource Allocation, Anomaly Detection, Dynamic Scaling

SmartSPIM Pipeline: A Scalable Cloud-Based Image Processing Pipeline for Light-sheet Microscopy Data

SmartSPIM Pipeline: A Scalable Cloud-Based Image Processing Pipeline for Light-sheet Microscopy Data

A Nursing Clinical Care Services Platform to Support Leadership Decision Making.

Like most organizations, The Children's Hospital of Philadelphia (CHOP) generates a lot of data of varying types, from clinical to staffing and operations. However, these datasets are separated, preventing effective analytics. To improve workforce management and care delivery, CHOP's Nursing department collaborated with the Data & Analytics team to create a dashboard-based platform for analyzing workforce metrics and patient outcomes, aiding nursing leaders in decision-making. This project involved multiple phases with contributions from Nursing, Data & Analytics, CHOP's Data Trust Office, and Human Resources. Key performance indicators were identified, and a variety of data sources were aggregated to provide a comprehensive view of the nursing enterprise. The resulting platform offers automated, current, and reliable analytics on various aspects, including nursing demographics, education, survey results, staffing actuals by job group, and patient and family experience data. The platform's usability was assessed using a modified Health Information Technology Usability Evaluation Scale survey, with a 29% response rate primarily from senior directors and managers. The findings showed high usability and satisfaction, indicating the dashboard is a valuable decision-support tool. Lessons learned include the need for analytics education for nurses and mid-managers, the inclusion of critical nursing-specific metrics (and development of the data pipelines making them possible), and the integration of multidisciplinary team metrics for comprehensive nursing analytics.

L-PBF High-Throughput Data Pipeline Approach for Multi-modal Integration

Metal-based additive manufacturing requires active monitoring solutions for assessing part quality. Multiple sensors and data streams, however, generate large heterogeneous data sets that are impractical for manual assessment and characterization. In this work, an automated pipeline is developed that enables feature extraction from high-speed camera video and multi-modal data analysis. The framework removes the need for manual assessment through the utilization of deep learning techniques and training models in a weakly supervised paradigm. We demonstrate this pipeline’s capability over 700,000 high-speed camera frames. The pipeline successfully extracts melt pool and spatter geometries and links them to corresponding pyrometry, radiography, and processparameter information. 715 individual prints are examined to reveal melt pool areas that exceeds 0.07 mm2 and pyrometry signal over a threshold (375 pyrometry units) were more likely to have defects. These automated processes enable massive throughput of characterization techniques.

Enhancing transparency in public procurement: A data-driven analytics approach

Open data is a strategy used by governments to promote transparency and accountability in public procurement processes. To reap the benefits of open data, exploring and analyzing the data is necessary to gain meaningful insights into procurement practices. However, accessing, processing, and analyzing open data can be challenging for non-data-savvy users with domain expertise, creating a barrier to leveraging open procurement data. To address this issue, we present the design, development, and implementation of a visual analytics tool. This tool automates data extraction from multiple sources, performs data cleansing, standardization, and database processing, and generates meaningful visualizations to streamline public procurement analysis. In addition, the tool estimates and visualizes corruption risk indicators at different levels (e.g., regions or public entities), providing valuable insights into the integrity of the procurement process. Key contributions of this work include: (1) providing a comprehensive guide to the development of an open procurement data visualization tool; (2) proposing a data pipeline to support processing, corruption risk estimator and data visualization; (3) demonstrating through a case study how visual analytics can effectively use open data to generate insights that promote and enhance transparency.

BioPipeline Creator—a user-friendly Java-based GUI for managing and customizing biological data pipelines

Bioinformatics tools are essential for performing analyses in the omics sciences. Given the numerous experimental opportunities arising from advances in the field of omics and easier access to high-throughput sequencing platforms, these tools play a fundamental role in research projects. Despite the considerable progress made possible by the development of bioinformatics tools, some tools are tailored to specific analytical goals, leading to challenges for non-bioinformaticians who need to integrate the results of these specific tools into a customized pipeline. To solve this problem, we have developed the BioPipeline Creator, a user-friendly Java-based GUI that allows different software tools to be integrated into the repertoire while ensuring easy user interaction via an accessible graphical interface. Consisting of client and server software components, BioPipeline Creator provides an intuitive graphical interface that simplifies the use of various bioinformatics tools for users without advanced computer skills. It can run on less sophisticated devices or workstations, allowing users to keep their operating system without having to switch to another compatible system. The server is responsible for the processing tasks and can perform the analysis in the user's local or remote network structure. Compatible with the most important operating systems, available at https://github.com/allanverasce/bpc.git.

High-Resolution Mass Spectrometry for Human Exposomics: Expanding Chemical Space Coverage.

In the modern "omics" era, measurement of the human exposome is a critical missing link between genetic drivers and disease outcomes. High-resolution mass spectrometry (HRMS), routinely used in proteomics and metabolomics, has emerged as a leading technology to broadly profile chemical exposure agents and related biomolecules for accurate mass measurement, high sensitivity, rapid data acquisition, and increased resolution of chemical space. Non-targeted approaches are increasingly accessible, supporting a shift from conventional hypothesis-driven, quantitation-centric targeted analyses toward data-driven, hypothesis-generating chemical exposome-wide profiling. However, HRMS-based exposomics encounters unique challenges. New analytical and computational infrastructures are needed to expand the analysis coverage through streamlined, scalable, and harmonized workflows and data pipelines that permit longitudinal chemical exposome tracking, retrospective validation, and multi-omics integration for meaningful health-oriented inferences. In this article, we survey the literature on state-of-the-art HRMS-based technologies, review current analytical workflows and informatic pipelines, and provide an up-to-date reference on exposomic approaches for chemists, toxicologists, epidemiologists, care providers, and stakeholders in health sciences and medicine. We propose efforts to benchmark fit-for-purpose platforms for expanding coverage of chemical space, including gas/liquid chromatography-HRMS (GC-HRMS and LC-HRMS), and discuss opportunities, challenges, and strategies to advance the burgeoning field of the exposome.

Consensus for Operating Room Multimodal Data Management: Identifying Research Priorities for Data-Driven Surgery.

This study aimed to identify research areas that demand attention in multimodal data-driven surgery for improving data management in minimally invasive surgery. New surgical procedures, high-tech equipment, and digital tools are increasingly being introduced, potentially benefiting patients and surgical teams. These innovations have resulted in operating rooms evolving into data-rich environments, which, in turn, requires a thorough understanding of the data pipeline for improved and more intelligent real-time data usage. As this new domain is vast, it is necessary to identify where efforts should be focused on developing seamless and practical data usage. A modified electronic Delphi approach was used; 53 investigators were divided into the following groups: a research group (n=9) for problem identification and a narrative literature review, a medical and technical expert group (n=14) for validation, and an invited panel (n=30) for two electronic survey rounds. Round 1 focused on a consensus regarding bottlenecks in surgical data science areas and research gaps, while round 2 prioritized the statements from round 1, and a roadmap was created based on the identified essential and very important research gaps. Consensus panelists have identified key research areas, including digitizing operating room (OR) activities, improving data streaming through advanced technologies, uniform protocols for handling multimodal data, and integrating AI for efficiency and safety. The roadmap prioritizes standardizing OR data formats, integrating OR data with patient information, ensuring regulatory compliance, standardizing surgical AI models, and securing data transfers in the next generation of wireless networks. This work is an international expert consensus regarding the current issues and key research targets in the promising field of data-driven surgery, highlighting the research needs of many operating room stakeholders with the aim of facilitating the implementation of novel patient care strategies in minimally invasive surgery.

Ontology‐Based Data Acquisition, Refinement, and Utilization in the Development of a Multilayer Ferrite Inductor

A key aspect in the development of multilayer inductors is the magnetic permeability of the ferrite layers. Herein, the effects of different processing steps on the permeability of a NiCuZn ferrite are investigated. Dry‐pressed, tape‐cast, and cofired multilayer samples are analyzed. An automated data pipeline is applied to structure the acquired experimental data according to a domain ontology based on Platform MaterialDigital core ontology. Example queries to the ontology show how the determined process–property correlations are accessible to nonexperts and thus how suitable data for component design can be identified. It is demonstrated how the inductance of cofired multilayer inductors is reliably predicted by simulations if the appropriate input data corresponding to the manufacturing process is used.

Evaluating the Reliability of a Social Presence Composite Construct for Online Computer Science Degree Programmes

As a contribution to ongoing discussions about the implications of social presence for online instruction technology, this study evaluated the reliability and validity of a composite social presence construct in online computer science programmes using archival data from the Computing Research Association’s (CRA) Center for Evaluating the Research Pipeline (CERP) Data Buddies Survey. Questions from the survey were aligned to three interdependent subconstructs of social presence proposed by Kreijns et al (2021). Spearman’s rank correlation coefficient was used to analyse the relationships between the subconstructs. Cronbach’s alpha was used to evaluate the internal consistency and reliability of the subconstructs of social presence, sociability, and social space, as well as the composite construct of social presence. The findings indicate that the social presence construct and its subconstructs are internally consistent and highly reliable, aligned with the CERP survey. Results indicated that the subconstructs are interrelated indicators of the perception of social presence in online computer science programmes. This study contributes to the literature concerning measuring social presence in online learning by providing a reliable and valid construct that can be used to assess the construct using different permutations of analysis on the CERP dataset.

PHANGS-JWST: Data-processing Pipeline and First Full Public Data Release

The exquisite angular resolution and sensitivity of JWST are opening a new window for our understanding of the Universe. In nearby galaxies, JWST observations are revolutionizing our understanding of the first phases of star formation and the dusty interstellar medium. Nineteen local galaxies spanning a range of properties and morphologies across the star-forming main sequence have been observed as part of the PHANGS-JWST Cycle 1 Treasury program at spatial scales of ∼5–50 pc. Here, we describe pjpipe, an image-processing pipeline developed for the PHANGS-JWST program that wraps around and extends the official JWST pipeline. We release this pipeline to the community as it contains a number of tools generally useful for JWST NIRCam and MIRI observations. Particularly for extended sources, pjpipe products provide significant improvements over mosaics from the MAST archive in terms of removing instrumental noise in NIRCam data, background flux matching, and calibration of relative and absolute astrometry. We show that slightly smoothing F2100W MIRI data to 0.″9 (degrading the resolution by about 30%) reduces the noise by a factor of ≈3. We also present the first public release (DR1.1.0) of the pjpipe processed eight-band 2–21 μm imaging for all 19 galaxies in the PHANGS-JWST Cycle 1 Treasury program. An additional 55 galaxies will soon follow from a new PHANGS-JWST Cycle 2 Treasury program.