Framework For Development Of Models Research Articles

Developing a Framework for Safe AI Model Development on Sensitive Healthcare Data

The Dementias Platform UK (DPUK) Data Portal holds over 60 cohort datasets from over 3 million participants with a range of multi-modal data including neuroimaging and genomics. This has meant we have seen an increasing interest in the development of AI models with the potential for clinical implementation. However, this presents a unique challenge to disclosure control when it comes to assessing these AI models for release due to the risk of attacks such as membership inference, model inversion or even just vulnerabilities in the models like overfitting. This is why we hosted a series of workshops bringing together members of the public, expert researchers, and data owners across the UK to build a framework for allowing the safe development and release of AI models trained on sensitive healthcare data. From this, we have put together recommendations and guidelines for the use of privacy-preserving techniques in AI models to protect patient privacy, and to allow the safe deployment of these models outside of trusted research environments. We also identified the unique challenges to privacy and implementation of privacy-preserving techniques in AI models regarding the use of complex data such as neuroimaging and genomics. This framework has created a path forward for supporting safe AI model development which takes into consideration rapidly evolving ethical, legal and privacy considerations.

Penerapan Algoritma Naïve Bayes Untuk Klasifikasi Status Gizi Stunting Pada Balita

Child stunting is a major public health concern in Indonesia. This study uses the Naïve Bayes classification algorithm to assess the nutritional condition of stunted children based on demographic and anthropometric characteristics. The information used comes from the Toddler Weighing Month (Bulan Penimbangan Balita - BPB) in Majalengka Regency. Data type conversion, separating data into training and testing sets, and data normalization are all examples of preprocessing steps. The model's evaluation results reveal an accuracy of 94.65%, with precision and recall for each category of stunted nutritional status. This study makes a substantial contribution to early diagnosis and mitigation of stunting in Indonesia, as well as providing the framework for future development of more powerful predictive models.

Data-driven visual model development and 3D visual analytics framework for underground mining

Data visualisation is broadly utilised to aid in presenting complex information, facilitating decision-making, and enhancing operational efficiency in the mining sector. However, due to the extensive utilisation of CAD files and the complexity and difficulty of converting 2D drawings into 3D models, it is challenging to realise (near) real-time 3D data visualisation in mining applications. The difficulties are including site drawing complexity, lack of standardisation, time-consuming cleaning, and imperative manual intervention. This paper addresses these challenges by focusing on an automatic, data-driven approach for 3D visual model development in mining, with an emphasis on lifetime support and sustainability. Therefore, we introduce standardised visual model data diagram and proposed conversion workflows to leverage digitised 2D drawings in automatic 3D visual model development. Moreover, to improve visualisation efficiency and facilitate visual analytics construction in mining, this paper also proposes a 3D tile index methodology for efficient visual model reconstruction and generic production data integration. To ensure lifetime support and sustainable development in visualisation, we propose two model expansion schemes: one based on dxf file-driven model development, and the other one is interaction-oriented 3D model design and development, which the two schemes share the same data exchange and conversion workflows. Furthermore, to facilitate data fusion among various datasets, we also provide insights into revealing spatial–temporal patterns among multimodality data by real-time visualisation. These advancements aim to propel the field of underground mining visualisation, offering significant insights into enhancing the efficiency and safety of mining operations.

Electromigration in Nano-Interconnects: Determining Reliability Margins in Redundant Mesh Networks Using a Scalable Physical-Statistical Hybrid Paradigm.

This paper presents a hybrid modelling approach that combines physics-based electromigration modelling (PEM) and statistical methods to evaluate the electromigration (EM) limits of nano-interconnects in mesh networks. The approach, which is also compatible with standard Place and Route (P&R) tools and practises, takes into account the positive impact of network redundancy on EM current limits. The numerical simulations conducted in this study show that conventional methods underestimate the EM current limits of a power delivery network (PDN) unit-cell by 80% due to their lack of consideration for redundancy. Additionally, the time-to-failure (TTF) distributions of a PDN unit-cell obtained by the developed modelling framework adhered to a lognormal distribution, where the lognormal sigma, σlogn, exhibits a 55% reduction compared to that of the single constituent interconnects. The study also found the negative voltage (i.e., ground or Vss) grid to be more susceptible to EM than the positive voltage, i.e., Vdd grid. In the examined grid unit-cell design, both the number of interconnect sites prone to voiding and also the magnitude of the peak tensile stress within the nano-interconnects were found to be two times as high in the Vss case compared to Vdd. The lognormal sigma of TFF for the grid unit-cells, σlogn-tile, show a marked reduction compared to the lognormal sigma of the constituent single interconnects, σlogn, with a 50% and 66% decrease compared to single interconnects, for downstream (Vss) and upstream (Vdd), respectively. In addition, σlogn-tile was three times higher for downstream (Vss) compared to upstream (Vdd), whilst, in contrast, this difference was only 2-fold at the single interconnect level. TTF50% was predicted to be 4.13-fold higher at the grid unit-cell level for the upstream compared to downstream operation, which was also more pronounced than in the single interconnect level where the difference was only 2-fold. This research provides valuable insights into the EM ageing of nano-interconnects in mesh networks and could pragmatically enhance the accuracy of EM compliance evaluation methods.

Uncertainty-aware output feedback model predictive combustion control of RCCI engines

Accurate model development is essential for effective model-based control of Reactivity Controlled Compression Ignition (RCCI) engines. However, due to the intricate nature of engine combustion process, achieving a precise model that can capture the complex dynamic behavior and ensure high control performance poses a significant challenge. In this paper, we propose an uncertainty-aware output feedback model predictive control approach for efficient combustion management in RCCI engines. In contrast to the previously developed approaches, this method adopts a data-driven approach within the linear parameter-varying (LPV) framework for model development. To address the model mismatch between the LPV model and the real system/data, Bayesian Neural Networks (BNNs) are employed which provide the probability distribution of the uncertainties. The BNNs enable the formation of a scenario tree, effectively characterizing the range of potential uncertainties in the system. Through the implementation of scenario-based model predictive control, our approach ensures high tracking performance for the RCCI engine in the presence of modeling uncertainties and measurement noise. Extensive simulations and experimental validations demonstrate the superiority of our uncertainty-aware model predictive control over traditional control strategies.

CausalOps — Towards an industrial lifecycle for causal probabilistic graphical models

Context:Causal probabilistic graph-based models have gained widespread utility, enabling the modeling of cause-and-effect relationships across diverse domains. With their rising adoption in new areas, such as safety analysis of complex systems, software engineering, and machine learning, the need for an integrated lifecycle framework akin to DevOps and MLOps has emerged. Currently, such a reference for organizations interested in employing causal engineering is missing. This lack of guidance hinders the incorporation and maturation of causal methods in the context of real-life applications. Objective:This work contextualizes causal model usage across different stages and stakeholders and outlines a holistic view of creating and maintaining them within the process landscape of an organization. Methods:A novel lifecycle framework for causal model development and application called CausalOps is proposed. By defining key entities, dependencies, and intermediate artifacts generated during causal engineering, a consistent vocabulary and workflow model to guide organizations in adopting causal methods are established. Results:Based on the early adoption of the discussed methodology to a real-life problem within the automotive domain, an experience report underlining the practicability and challenges of the proposed approach is discussed. Conclusion:It is concluded that besides current technical advancements in various aspects of causal engineering, an overarching lifecycle framework that integrates these methods into organizational practices is missing. Although diverse skills from adjacent disciplines are widely available, guidance on how to transfer these assets into causality-driven practices still need to be addressed in the published literature. CausalOps’ aim is to set a baseline for the adoption of causal methods in practical applications within interested organizations and the causality community.

Analyzing coherent and incoherent optical susceptibilities in four-wave mixing: stochasticity, symmetry, and intramolecular coupling

The nonlinear optical susceptibilities are analyzed in their optical propagation length considering the incoherent-coherent contributions with the dependence on the relaxation times. In addition, some symmetry properties of the Four-Wave mixing (FWM) signal concerning the frequency detuning exchange are used in the study, without restricting ourselves to the maxima of population oscillations. We analyze the behaviors of these optical responses in a vibronic coupling scheme using a molecule model consisting of two coupled harmonic curves of electronic energies with displaced minima in position and nuclear energies. The solvent effect in our model is treated through the natural Bohr frequency shift to a time-dependent function, with explicit manifestations in its comparison as if the upper state were broadened. Our results for both FWM signal propagation and nonlinear optical susceptibilities are sensitive to the intramolecular coupling parameters, solvent stochasticity, and perturbation order for the treatment of the weak probe beam. The consideration of a second-order probe beam in FWM would not represent a significant contribution to the predicted results, since its percentage contribution is negligible within the model development framework; however, this contribution becomes important when increasing the ratio between longitudinal and transverse relaxation times.

A 3D In-vitro model of the human dentine interface shows long-range osteoinduction from the dentine surface

Emerging regenerative cell therapies for alveolar bone loss have begun to explore the use of cell laden hydrogels for minimally invasive surgery to treat small and spatially complex maxilla-oral defects. However, the oral cavity presents a unique and challenging environment for in vivo bone tissue engineering, exhibiting both hard and soft periodontal tissue as well as acting as key biocenosis for many distinct microbial communities that interact with both the external environment and internal body systems, which will impact on cell fate and subsequent treatment efficacy. Herein, we design and bioprint a facile 3D in vitro model of a human dentine interface to probe the effect of the dentine surface on human mesenchymal stem cells (hMSCs) encapsulated in a microporous hydrogel bioink. We demonstrate that the dentine substrate induces osteogenic differentiation of encapsulated hMSCs, and that both dentine and β-tricalcium phosphate substrates stimulate extracellular matrix production and maturation at the gel-media interface, which is distal to the gel-substrate interface. Our findings demonstrate the potential for long-range effects on stem cells by mineralized surfaces during bone tissue engineering and provide a framework for the rapid development of 3D dentine-bone interface models.

A comprehensive AI model development framework for consistent Gleason grading

BackgroundArtificial Intelligence(AI)-based solutions for Gleason grading hold promise for pathologists, while image quality inconsistency, continuous data integration needs, and limited generalizability hinder their adoption and scalability.MethodsWe present a comprehensive digital pathology workflow for AI-assisted Gleason grading. It incorporates A!MagQC (image quality control), A!HistoClouds (cloud-based annotation), Pathologist-AI Interaction (PAI) for continuous model improvement, Trained on Akoya-scanned images only, the model utilizes color augmentation and image appearance migration to address scanner variations. We evaluate it on Whole Slide Images (WSI) from another five scanners and conduct validations with pathologists to assess AI efficacy and PAI.ResultsOur model achieves an average F1 score of 0.80 on annotations and 0.71 Quadratic Weighted Kappa on WSIs for Akoya-scanned images. Applying our generalization solution increases the average F1 score for Gleason pattern detection from 0.73 to 0.88 on images from other scanners. The model accelerates Gleason scoring time by 43% while maintaining accuracy. Additionally, PAI improve annotation efficiency by 2.5 times and led to further improvements in model performance.ConclusionsThis pipeline represents a notable advancement in AI-assisted Gleason grading for improved consistency, accuracy, and efficiency. Unlike previous methods limited by scanner specificity, our model achieves outstanding performance across diverse scanners. This improvement paves the way for its seamless integration into clinical workflows.

A prognostic and health management approach using colour fade to determine the condition of silk in a museum display environment

Silk is highly susceptible to fade, often resulting in colour loss to the fabric even when displayed in controlled museum environments. This study applies the engineering-inspired Prognostics and Health Management (PHM) approach to assess the remaining useful life (RUL) of silk fabric in a museum environment, focussing on predicting colour fade as an indicator of degradation. A novel mathematical model is developed to forecast cumulative colour fade of silk upholstery exhibited at the Great Gallery in The Wallace Collection, London. Unlike traditional accelerated aging methods, this model utilises naturally aged samples to estimate the rate of colour change over time. The main contribution of this work lies in the methodology and computational framework for model development using environmental data and conditions of silk in the museum environment. The proposed PHM methodology allows for continuous modifications of the colour fade model to improve accuracy by deploying new data from different silk collections and environments. The article demonstrates a model-based approach for informed decision making in museums regarding display and storage of silk upholstery.

Sustainable management of land use patterns and water allocation for coordinated multidimensional development

The sustainable and integrated management of resources is a crucial approach to achieving sustainable regional development in the face of limited land and increasingly scarce water resources. This paper proposes a multidimensional optimization model to jointly optimize land use patterns and water resource allocation under the conditions of maximizing economic benefits and minimizing social differences, ecological loads, environmental pollution, and carbon emissions. The optimization model is solved using the affiliation function method and then combined with random forest (RF) and cellular automata (CA) models to further obtain the optimal spatial layout of land use. This paper takes the Mid-Yangtze River City Cluster as an example and optimizes the land use pattern and water allocation in the region in 2020 and 2030. The results of the study show that the optimized water and land resource allocation scheme performs better than the status quo in all dimensions. The optimized economic benefits, social effects, ecological effects, environmental effects, and overall sustainability in 2020 are increased by 45.97%, 22.89%, 10.07%, 24.65%, and 42.73%, respectively. By 2030, overall sustainability reaches 0.69. The developed modeling framework helps to elucidate the internal linkages between land use and water allocation, and it provides decision-makers with diversified approaches to land and water resource management and adjustment for different cities in different periods, which promotes the sustainable development of urban agglomerations.

Application of transfer learning to overcome data imbalance and extrapolation for model predictive control: A real-life case

This paper proposes a transfer learning (TL)-based control-oriented model development framework. In particular, this study examines the transferability from virtual (source) to existing (target) buildings to overcome the data imbalance issue of the data-driven approach. The target system is a cooling system comprising two supply air fans and four condensing units. First, synthetic data rich enough to provide fundamental knowledge about the target system were generated using the EnergyPlus model. A data-driven model was subsequently developed to learn the underlying dynamics of the system. By adopting TL using an imbalanced dataset measured from the target system, the knowledge that the model learned from the virtual data was transferred to the target system of the existing building. The results showed that the transfer learning model could accurately describe the dynamic behavior of the target system and predict the supply air temperature with marginal errors (CVRMSE: 5.4%, MAE: 0.96 ℃). In other words, the TL from virtual to existing buildings can overcome the data imbalance issue for developing a reliable data-driven model.

Understanding the influence of surgical parameters on craniofacial surgery outcomes: a computational study.

Sagittal craniosynostosis (SC) is a congenital condition whereby the newborn skull develops abnormally owing to the premature ossification of the sagittal suture. Spring-assisted cranioplasty (SAC) is a minimally invasive surgical technique to treat SC, where metallic distractors are used to reshape the newborn's head. Although safe and effective, SAC outcomes remain uncertain owing to the limited understanding of skull-distractor interaction and the limited information provided by the analysis of single surgical cases. In this work, an SC population-averaged skull model was created and used to simulate spring insertion by means of the finite-element analysis using a previously developed modelling framework. Surgical parameters were varied to assess the effect of osteotomy and spring positioning, as well as distractor combinations, on the final skull dimensions. Simulation trends were compared with retrospective measurements from clinical imaging (X-ray and three-dimensional photogrammetry scans). It was found that the on-table post-implantation head shape change is more sensitive to spring stiffness than to the other surgical parameters. However, the overall end-of-treatment head shape is more sensitive to spring positioning and osteotomy size parameters. The results of this work suggest that SAC surgical planning should be performed in view of long-term results, rather than immediate on-table reshaping outcomes.

Artificial Intelligence (AI) Model Development Framework for the Protection of State Borders, with a Focus on Analyzing Behavioral Patterns

This study explores the development and implementation of an artificial intelligence (AI) model designed to predict illegal border crossing locations, thereby enhancing the effectiveness of national border security measures. By integrating and analyzing diverse data sources, -including satellite imagery, social media, and environmental factors, -this AI model aims to identify potential migration patterns and high-risk areas for illegal crossing. This research highlights the model's ability to provide real-time risk assessments, offering a novel approach to border security that surpasses traditional methods in terms of both efficiency and cost-effectiveness. The model's adaptability, continuous learning capabilities, and user-friendly interfaces ensure its relevance in addressing contemporary border-security challenges. This article contributes to the ongoing discourse on the application of AI in national security by proposing a solution that leverages technology for improved coordination among immigration services, government organizations, and international bodies.

Continuous-Time Graph Representation with Sequential Survival Process

Over the past two decades, there has been a tremendous increase in the growth of representation learning methods for graphs, with numerous applications across various fields, including bioinformatics, chemistry, and the social sciences. However, current dynamic network approaches focus on discrete-time networks or treat links in continuous-time networks as instantaneous events. Therefore, these approaches have limitations in capturing the persistence or absence of links that continuously emerge and disappear over time for particular durations. To address this, we propose a novel stochastic process relying on survival functions to model the durations of links and their absences over time. This forms a generic new likelihood specification explicitly accounting for intermittent edge-persistent networks, namely GraSSP: Graph Representation with Sequential Survival Process. We apply the developed framework to a recent continuous time dynamic latent distance model characterizing network dynamics in terms of a sequence of piecewise linear movements of nodes in latent space. We quantitatively assess the developed framework in various downstream tasks, such as link prediction and network completion, demonstrating that the developed modeling framework accounting for link persistence and absence well tracks the intrinsic trajectories of nodes in a latent space and captures the underlying characteristics of evolving network structure.

Prediction of Effectiveness and Toxicities of Immune Checkpoint Inhibitors Using Real-World Patient Data.

Although immune checkpoint inhibitors (ICIs) have improved outcomes in certain patients with cancer, they can also cause life-threatening immunotoxicities. Predicting immunotoxicity risks alongside response could provide a personalized risk-benefit profile, inform therapeutic decision making, and improve clinical trial cohort selection. We aimed to build a machine learning (ML) framework using routine electronic health record (EHR) data to predict hepatitis, colitis, pneumonitis, and 1-year overall survival. Real-world EHR data of more than 2,200 patients treated with ICI through December 31, 2018, were used to develop predictive models. Using a prediction time point of ICI initiation, a 1-year prediction time window was applied to create binary labels for the four outcomes for each patient. Feature engineering involved aggregating laboratory measurements over appropriate time windows (60-365 days). Patients were randomly partitioned into training (80%) and test (20%) sets. Random forest classifiers were developed using a rigorous model development framework. The patient cohort had a median age of 63 years and was 61.8% male. Patients predominantly had melanoma (37.8%), lung cancer (27.3%), or genitourinary cancer (16.4%). They were treated with PD-1 (60.4%), PD-L1 (9.0%), and CTLA-4 (19.7%) ICIs. Our models demonstrate reasonably strong performance, with AUCs of 0.739, 0.729, 0.755, and 0.752 for the pneumonitis, hepatitis, colitis, and 1-year overall survival models, respectively. Each model relies on an outcome-specific feature set, though some features are shared among models. To our knowledge, this is the first ML solution that assesses individual ICI risk-benefit profiles based predominantly on routine structured EHR data. As such, use of our ML solution will not require additional data collection or documentation in the clinic.

Evaluation of efferent influences on neural coding using pre-clinical models of sensorineural hearing loss

The medial olivocochlear (MOC) efferent system is less explored than the ascending auditory pathway but likely contributes in important ways to neural coding and perception. These effects are thought to vary across stimulus configurations, anesthetic states, and subtypes of sensorineural hearing loss (SNHL). To explore effective assays of MOC effects on neural coding, we have recorded Interleaved otoacoustic emissions (OAEs) and envelope following responses (EFRs) from several pre-clinical SNHL chinchilla models. Our preliminary observations include increases in OAEs with inner-hair-cell loss and anesthesia, which may be due to reduced efferent strength. Additionally, we observed enhanced EFRs with an added noise masker, which also could be related to efferent effects. We are using sedated and awake OAE comparisons to develop a standard efferent assay for use in neural-coding studies. Interleaved recording of OAEs and EFRs track cochlear-gain and neural-coding changes during acoustic stimuli. A recently developed modeling framework (Farhadi et al., 2023 JASA) that includes different MOC projection pathways, including midbrain modulation-sensitive inputs, is used to guide most-effective stimulus selection. Ultimately, this model-guided experimental framework will provide unique guidance for testing MOC hypotheses related to neural coding.

Unsupervised deep learning framework for temperature-compensated damage assessment using ultrasonic guided waves on edge device

Fueled by the rapid development of machine learning (ML) and greater access to cloud computing and graphics processing units, various deep learning based models have been proposed for improving performance of ultrasonic guided wave structural health monitoring (GW-SHM) systems, especially to counter complexity and heterogeneity in data due to varying environmental factors (e.g., temperature) and types of damages. Such models typically comprise of millions of trainable parameters, and therefore add to cost of deployment due to requirements of cloud connectivity and processing, thus limiting the scale of deployment of GW-SHM. In this work, we propose an alternative solution that leverages TinyML framework for development of light-weight ML models that could be directly deployed on embedded edge devices. The utility of our solution is illustrated by presenting an unsupervised learning framework for damage detection in honeycomb composite sandwich structure with disbond and delamination type of damages, validated using data generated by finite element simulations and experiments performed at various temperatures in the range 0–90 °C. We demonstrate a fully-integrated solution using a Xilinx Artix-7 FPGA for data acquisition and control, and edge-inference of damage. Despite the limited number of features, the lightweight model shows reasonably high accuracy, thereby enabling detection of small size defects with improved sensitivity on an edge device for online GW-SHM.

The XSO framework (v0.1) and Phydra library (v0.1) for a flexible, reproducible, and integrated plankton community modeling environment in Python

Abstract. Plankton community modeling is a critical tool for understanding the processes that shape marine ecosystems and their impacts on global biogeochemical cycles. These models can be of variable ecological, physiological, and physical complexity. Many published models are either not publicly available or implemented in static and inflexible code, thus hampering adoption, collaboration, and reproducibility of results. Here we present Phydra, an open-source library for plankton community modeling, and Xarray-simlab-ODE (XSO), a modular framework for efficient, flexible, and reproducible model development based on ordinary differential equations. Both tools are written in Python. Phydra provides pre-built models and model components that can be modified and assembled to develop plankton community models of various levels of ecological complexity. The components can be created, adapted, and modified using standard variable types provided by the XSO framework. XSO is embedded in the Python scientific ecosystem and is integrated with tools for data analysis and visualization. To demonstrate the range of applicability and how Phydra and XSO can be used to develop and execute models, we present three applications: (1) a highly simplified nutrient–phytoplankton (NP) model in a chemostat setting, (2) a nutrient–phytoplankton–zooplankton–detritus (NPZD) model in a zero-dimensional pelagic ocean setting, and (3) a size-structured plankton community model that resolves 50 phytoplankton and 50 zooplankton size classes with functional traits determined by allometric relationships. The applications presented here are available as interactive Jupyter notebooks and can be used by the scientific community to build, modify, and run plankton community models based on differential equations for a diverse range of scientific pursuits.

A Predictive Model of Vaccine Reactogenicity Using Data from an InVitro Human Innate Immunity Assay System.

A primary concern in vaccine development is safety, particularly avoiding an excessive immune reaction in an otherwise healthy individual. An accurate prediction of vaccine reactogenicity using invitro assays and computational models would facilitate screening and prioritization of novel candidates early in the vaccine development process. Using the modular invitro immune construct model of human innate immunity, PBMCs from 40 healthy donors were treated with 10 different vaccines of varying reactogenicity profiles and then cell culture supernatants were analyzed via flow cytometry and a multichemokine/cytokine assay. Differential response profiles of innate activity and cell viability were observed in the system. In parallel, an extensive adverse event (AE) dataset for the vaccines was assembled from clinical trial data. A novel reactogenicity scoring framework accounting for the frequency and severity of local and systemic AEs was applied to the clinical data, and a machine learning approach was employed to predict the incidence of clinical AEs from the invitro assay data. Biomarker analysis suggested that the relative levels of IL-1B, IL-6, IL-10, and CCL4 have higher predictive importance for AE risk. Predictive models were developed for local reactogenicity, systemic reactogenicity, and specific individual AEs. A forward-validation study was performed with a vaccine not used in model development, Trumenba (meningococcal group B vaccine). The clinically observed Trumenba local and systemic reactogenicity fell on the 26th and 93rd percentiles of the ranges predicted by the respective models. Models predicting specific AEs were less accurate. Our study presents a useful framework for the further development of vaccine reactogenicity predictive models.