Combining Machine Learning Research Articles

Altered dynamic large-scale brain networks and combined machine learning in primary angle-closure glaucoma

Primary angle-closure glaucoma (PACG) is a severe and irreversible blinding eye disease characterized by progressive retinal ganglion cell death. However, prior research has predominantly focused on static brain activity changes, neglecting the exploration of how PACG impacts the dynamic characteristics of functional brain networks. This study enrolled forty-four patients diagnosed with PACG and forty-four age, gender, and education level-matched healthy controls (HCs). The study employed Independent Component Analysis (ICA) techniques to extract resting-state networks (RSNs) from resting-state functional magnetic resonance imaging (rs-fMRI) data. Subsequently, the RSNs was utilized as the basis for examining and comparing the functional connectivity variations within and between the two groups of resting-state networks. To further explore, a combination of sliding time window and k-means cluster analyses identified seven stable and repetitive dynamic functional network connectivity (dFNC) states. This approach facilitated the comparison of dynamic functional network connectivity and temporal metrics between PACG patients and HCs for each state. Subsequently, a support vector machine (SVM) model leveraging functional connectivity (FC) and FNC was applied to differentiate PACG patients from HCs. Our study underscores the presence of modified functional connectivity within large-scale brain networks and abnormalities in dynamic temporal metrics among PACG patients. By elucidating the impact of changes in large-scale brain networks on disease evolution, researchers may enhance the development of targeted therapies and interventions to preserve vision and cognitive function in PACG.

Machine learning-based discrimination of benign and malignant breast lesions on US: The contribution of shear-wave elastography

PurposeTo build and validate a combined radiomics and machine learning (ML) approach using B-mode US and SWE images to differentiate benign from malignant solid breast lesions (BLs) and compare its performance with that of an expert radiologist. MethodsPatients with at least one BI-RADS 2–6 BL who performed breast US integrated with SWE were retrospectively included. B-mode US and SWE images were manually segmented to extract radiomics features. A multi-step feature selection process was performed and a predictive model built using the Logistic Regression algorithm. The diagnostic accuracy was evaluated with the AUC and Matthews Correlation Coefficient (MCC) metrics. The performance of the ML classifier was compared to that of an expert radiologist. Results427 Bls were included and divided into a training (286 BLs, of which 127 benign and 159 malignant) and a test set (141 BLs, of which 59 benign and 82 malignant). Of 1098 features extracted from B-mode US and SWE images, 13 were finally selected. The ML classifier showed an AUC of 0.768 and 0.746, and an MCC of 0.403 and 0.423 in the training and test sets, respectively. The performance was higher than that of the expert radiologist assessing only B-mode US images, but significantly lower when SWE images were also provided. ConclusionA ML approach based on B-mode US and SWE images may represent a potential tool in the characterization of BLs. SWE still gives its most relevant contribution in the clinical setting rather than included in a radiomics pipeline.

Searching for potential acetylcholinesterase inhibitors: a combined approach of multi-step similarity search, machine learning and molecular dynamics simulations.

Targeting acetylcholinesterase is one of the most important strategies for developing therapeutics against Alzheimer's disease. In this work, we have employed a new approach that combines machine learning models, a multi-step similarity search of the PubChem library and molecular dynamics simulations to investigate potential inhibitors for acetylcholinesterase. Our search strategy has been shown to significantly enrich the set of compounds with strong predicted binding affinity to acetylcholinesterase. Both machine learning prediction and binding free energy calculation, based on linear interaction energy, suggest that the compound CID54414454 would bind strongly to acetylcholinesterase and hence is a promising inhibitor.

Forecasting CO2 Emissions from Libya’s Transport Sector

This paper presents an innovative approach to forecast carbon dioxide (CO2) emissions from the transport sector in Libya. The method combines machine learning algorithms with historical data and future estimates. The research built a model that took into account factors such as population growth, rates of car ownership, patterns of fuel consumption and government regulations in order to provide an accurate forecast of carbon dioxide (CO2) emissions over the next decade based on the Global Change Assessment Model (GCAM). The authors used a variety of statistical time series models to forecast future CO2 emissions from Libya's transportation sector. These models included the exponential smoothing model (ESM) and the autoregressive integrated moving average (ARIMA). The ARIMA model outperformed the ESM model, achieving an R2 of 0.931 and a root mean square error (RMSE) of 1.040 Mt CO2. The results of the study found that CO2 emissions from Libya's transport sector could increase by 27.98% and 57.99% in 2030 and 2050, respectively. The study proposed six transportation theories to reduce CO2 emissions from Africa's and Libya's transport sectors. The identified factors encompass price systems, land use planning, eco-driving, electric automobiles, bicycle infrastructure, and telecommuting. The authors also examined the needs to reduce CO2 emissions from Libya’s transport sector in order to meet the International Energy Agency’s ambitious targets for reducing CO2 emissions from the global transport sector. These needs arise due to increasing urbanization, population growth, underinvestment in public transportation infrastructure, and the increasing incidence and severity of heat waves. Additionally, hypothetical scenarios are presented to demonstrate the importance of further reducing CO2 emissions from these sectors to match the projections of global change assessment models.

Data-driven multi-objective optimization of aerodynamics and aeroacoustics in dual Savonius wind turbines using large eddy simulation and machine learning

Savonius rotor is a popular form of vertical-axis wind turbine (VAWT) for small-scale and urban applications because of its straightforward design and self-starting ability. Dual VAWTs present challenges in terms of wake interactions and noise, particularly in urban areas. Optimizing these parameters is essential for future wind energy adoption. This research is the first to analyze how the interaction of wakes from adjacent rotors, combined with a deflector, affects both the aerodynamic performance and noise levels of dual Savonius rotors. Large Eddy Simulation is applied, as it effectively captures detailed turbulent wind flows and their interactions with wind turbines. A multi-objective optimization method combining Machine Learning and Computational Fluid Dynamics (CFD) is developed to optimize rotors for maximum power efficiency and minimum noise, considering their wake interactions with a unique deflector system. First, the influence of geometric parameters on aerodynamics and aeroacoustics characteristics of rotors is analyzed, and the database is generated using Design of Experiment approach. Next, the CFD model is replaced by Artificial Neural Network (ANN) model established for predicting rotor performances. A Multi-Objective Genetic Algorithm method is used to optimize aerodynamics and aeroacoustics characteristics of rotors. Finally, optimal design parameters are identified from the Pareto front using the technique for order of preference by similarity to ideal solution decision-making method. The ANN model demonstrated high accuracy with an RANN2 of 0.995 and 0.971 for the average power coefficient (CP) and overall sound pressure level (OSPL) predictions, respectively. Multi-objective optimization revealed the best configuration of the deflector with bleed jets, improving the average CP up to 57.5% and reducing OSPL to an almost 5.2% compared to the dual rotor case at TSR = 0.8.

Combining machine learning and molecular dynamics to predict strength-toughness and energy dissipation mechanisms of hybrid double-crosslinked CNT networks

Cross-linking has a significant strengthening effect on the mechanical properties of carbon nanoparticles, but there are still great challenges in how to control their mechanical properties through precise cross-linking ratios. Therefore, we use coarse-grained molecular dynamics (CGMD) to simulate its mechanical properties as data samples and combine it with machine learning methods to predict the optimal strength and toughness of carbon nanotubes (CNTs) corresponding to optimal cross-linking conditions. We found that the cross-linking density range is 2.60 ∼ 2.75 × 10-4/nm3, and the strong and weak cross-linking ratio range is 91 % S+9% W∼95 % S+5% W, which is the optimal range to achieve the mechanical properties of CNTs. Excessively high cross-link density and strong bond ratios can reduce the total number of cross-link bond breaks, thus decreasing their toughness. Meanwhile, high cross-linking density has a greater impact on the shear and slip between CNTs, tending to make CNTs brittle. Our proposed new approach for extracting data from coarse-grained models can obtain substantial optimization results without requiring much simulation computation. Furthermore, these cross-linking strategies and machine learning algorithms proposed in this work can provide a theoretical basis for controlling the mechanical properties of CNT materials, and also open up a new way for functional network materials other than CNTs.

A forming method of variable-diameter tubes straightforward from sheet metals

Variable-diameter metallic tubes are extensively used in many industrial applications. They are generally made of pre-formed raw tubes by using the existing processing methods, during which obvious non-uniformity in tube wall thickness inevitably occurs. In this work, a novel method is proposed to form variable-diameter tubes directly from sheet metals by means of the roll-stamp forming technique. A three-pass roll-stamp forming process and corresponding die blocks are designed and implemented to form a demonstrative conical tube. The finite element model of the roll-stamp forming process is established and forming quality is quantitatively evaluated. The initial blank configuration, which is the most important factor for process design, is determined by combining machine learning and multi-objective optimization. The forming experiment is performed, and the conical tube is successfully formed with no significant thinning/thickening of tube wall. The influences of forming direction and conical ratio on the forming process are numerically investigated. The proposed roll-stamp technique provides a novel and effective approach to manufacturing variable-diameter tubes.

Combining Machine Learning and Remote Sensing for Online Environmental Quality Monitoring and Cause Analysis in Xiamen, China

Combining Machine Learning and Remote Sensing for Online Environmental Quality Monitoring and Cause Analysis in Xiamen, China

A novel deep domain adaptation method based on three-way decisions for identifying autistic patients

Combining machine learning with functional magnetic resonance imaging (fMRI) technology to build an effective autism recognition model has become a research hotspot. However, the performance of existing machine learning models is highly dependent on large-scale labelled data, and the cost of fMRI data acquisition and labelling limits the application of models in autism recognition. Although the availability of multisite data increases the sample size, the resolution, image contrast, and noise level of fMRI data acquired at different sites vary considerably due to the different scanning devices or scanning protocols used. The heterogeneity of fMRI data between sites leads to limited generalisation of the trained models. To tackle this problem, we propose a deep domain adaptation approach based on three-way decisions. First, we introduce a pseudolabel optimisation method based on three-way decisions, which fully considers the structural information of the target domain and makes secondary decisions on the uncertain objects in the target domain, thus improving the accuracy of the pseudolabel in the target domain. Then, we construct a new autism recognition network using a stacked autoencoder and combine the proposed pseudolabel optimisation model to adapt the domain shift. Finally, the self-training strategy is used to retrain the network after domain adaptation to improve the generalisability of the model to the target domain. Experimental results on several cross-domain autism recognition tasks show the effectiveness of the proposed method.

Large-scale deep learning identifies the antiviral potential of PKI-179 and MTI-31 against coronaviruses

Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) has led to the global pandemic of Coronavirus Disease (2019) (COVID-19), underscoring the urgency for effective antiviral drugs. Despite the development of different vaccination strategies, the search for specific antiviral compounds remains crucial. Here, we combine machine learning (ML) techniques with in vitro validation to efficiently identify potential antiviral compounds. We overcome the limited amount of SARS-CoV-2 data available for ML using various techniques, supplemented with data from diverse biomedical assays, which enables end-to-end training of a deep neural network architecture. We use its predictions to identify and prioritize compounds for in vitro testing. Two top-hit compounds, PKI-179 and MTI-31, originally identified as Pi3K-mTORC1/2 pathway inhibitors, exhibit significant antiviral activity against SARS-CoV-2 at low micromolar doses. Notably, both compounds outperform the well-known mTOR inhibitor rapamycin. Furthermore, PKI-179 and MTI-31 demonstrate broad-spectrum antiviral activity against SARS-CoV-2 variants of concern and other coronaviruses. In a physiologically relevant model, both compounds show antiviral effects in primary human airway epithelial (HAE) cultures derived from healthy donors cultured in an air-liquid interface (ALI). This study highlights the potential of ML combined with in vitro testing to expedite drug discovery, emphasizing the adaptability of AI-driven approaches across different viruses, thereby contributing to pandemic preparedness.

In Silico Modeling of Fabry Disease Pathophysiology for the Identification of Early Cellular Damage Biomarker Candidates.

Fabry disease (FD) is an X-linked lysosomal disease whose ultimate consequences are the accumulation of sphingolipids and subsequent inflammatory events, mainly at the endothelial level. The outcomes include different nervous system manifestations as well as multiple organ damage. Despite the availability of known biomarkers, early detection of FD remains a medical need. This study aimed to develop an in silico model based on machine learning to identify candidate vascular and nervous system proteins for early FD damage detection at the cellular level. A combined systems biology and machine learning approach was carried out considering molecular characteristics of FD to create a computational model of vascular and nervous system disease. A data science strategy was applied to identify risk classifiers by using 10 K-fold cross-validation. Further biological and clinical criteria were used to prioritize the most promising candidates, resulting in the identification of 36 biomarker candidates with classifier abilities, which are easily measurable in body fluids. Among them, we propose four candidates, CAMK2A, ILK, LMNA, and KHSRP, which have high classification capabilities according to our models (cross-validated accuracy ≥ 90%) and are related to the vascular and nervous systems. These biomarkers show promise as high-risk cellular and tissue damage indicators that are potentially applicable in clinical settings, although in vivo validation is still needed.

PROTA: A Robust Tool for Protamine Prediction Using a Hybrid Approach of Machine Learning and Deep Learning

Protamines play a critical role in DNA compaction and stabilization in sperm cells, significantly influencing male fertility and various biotechnological applications. Traditionally, identifying these proteins is a challenging and time-consuming process due to their species-specific variability and complexity. Leveraging advancements in computational biology, we present PROTA, a novel tool that combines machine learning (ML) and deep learning (DL) techniques to predict protamines with high accuracy. For the first time, we integrate Generative Adversarial Networks (GANs) with supervised learning methods to enhance the accuracy and generalizability of protamine prediction. Our methodology evaluated multiple ML models, including Light Gradient-Boosting Machine (LIGHTGBM), Multilayer Perceptron (MLP), Random Forest (RF), eXtreme Gradient Boosting (XGBOOST), k-Nearest Neighbors (KNN), Logistic Regression (LR), Naive Bayes (NB), and Radial Basis Function-Support Vector Machine (RBF-SVM). During ten-fold cross-validation on our training dataset, the MLP model with GAN-augmented data demonstrated superior performance metrics: 0.997 accuracy, 0.997 F1 score, 0.998 precision, 0.997 sensitivity, and 1.0 AUC. In the independent testing phase, this model achieved 0.999 accuracy, 0.999 F1 score, 1.0 precision, 0.999 sensitivity, and 1.0 AUC. These results establish PROTA, accessible via a user-friendly web application. We anticipate that PROTA will be a crucial resource for researchers, enabling the rapid and reliable prediction of protamines, thereby advancing our understanding of their roles in reproductive biology, biotechnology, and medicine.

A machine learning and linear programming aided approach to wine ranking and selection

PurposeThis paper aims to mitigate the subjective nature of wine rating by introducing statistical and optimization tools for analysis, providing a unique approach not found in existing literature.Design/methodology/approachThe research uses an unsupervised machine learning algorithm, k-means, to cluster wines based on their chemical characteristics, followed by the application of the PROMETHEE II multicriteria decision-making model to rank the wines based on their sensorial characteristics and selling price. Lastly, a linear programming model is used to optimize the selection of wines under different scenarios and constraints.FindingsThe study presents a method to rank wines based on both chemical and sensorial characteristics, providing a more comprehensive assessment than traditional subjective ratings. Clustering wines based on their characteristics and ranking them according to sensorial characteristics provides the user/consumer with meaningful information to be used in an optimization model for wine selection.Practical implicationsThe proposed framework has practical implications for wine enthusiasts, makers, tasters and retailers, offering a systematic approach to ranking and selecting/recommending wines based on both objective and subjective criteria. This approach can influence pricing, consumption and marketing strategies within the wine industry, leading to more informed and precise decision-making.Originality/valueThe research introduces a novel framework that combines machine learning, decision-making models and linear programming for wine ranking and selection, addressing the limitations of subjective ratings and providing a more objective approach.

Which perceptual categories do observers experience during multistable perception?

Multistable perceptual phenomena provide insights into the mind's dynamic states within a stable external environment and the neural underpinnings of these consciousness changes are often studied with binocular rivalry. Conventional methods to study binocular rivalry suffer from biases and assumptions that limit their ability to describe the continuous nature of this perceptual transitions and to discover what kind of percept was perceived across time. In this study, we propose a novel way to avoid those shortcomings by combining a continuous psychophysical method that estimates introspection during binocular rivalry with machine learning clustering and transition probability analysis. This combination of techniques reveals individual variability and complexity of perceptual experience in 28 normally sighted participants. Also, the analysis of transition probabilities between perceptual categories, i.e., exclusive and different kinds of mixed percepts, suggest that interocular perceptual competition, triggered by low-level stimuli, involves conflict between monocular and binocular neural processing sites rather than mutual inhibition of monocular sites.

Machine learning and domain decomposition methods - a survey

Hybrid algorithms, which combine black-box machine learning methods with experience from traditional numerical methods and domain expertise from diverse application areas, are progressively gaining importance in scientific machine learning and various industrial domains, especially in computational science and engineering. In the present survey, several promising avenues of research will be examined which focus on the combination of machine learning (ML) and domain decomposition methods (DDMs). The aim of this survey is to provide an overview of existing work within this field and to structure it into domain decomposition for machine learning and machine learning-enhanced domain decomposition, including: domain decomposition for classical machine learning, domain decomposition to accelerate the training of physics-aware neural networks, machine learning to enhance the convergence properties or computational efficiency of DDMs, and machine learning as a discretization method in a DDM for the solution of PDEs. In each of these fields, we summarize existing work and key advances within a common framework and, finally, discuss ongoing challenges and opportunities for future research.

A machine-learning model of academic resilience in the times of the COVID-19 pandemic: Evidence drawn from 79 countries/economies in the PISA 2022 mathematics study.

Given that students from socio-economically disadvantaged family backgrounds are more likely to suffer from low academic performance, there is an interest in identifying features of academic resilience, which may mitigate the relationship between disadvantaged socio-economic status and academic performance. This study sought to combine machine learning and explainable artificial intelligence (XAI) technique to identify key features of academic resilience in mathematics learning during COVID-19. Based on PISA 2022 data in 79 countries/economies, the random forest model coupled with Shapley additive explanations (SHAP) value technique not only uncovered the key features of academic resilience but also examined the contributions of each key feature. Findings indicated that 35 features were identified in the classification of academically resilient and non-academically resilient students, which largely validated the previous academic resilient framework. Notably, gender differences were shown in the distribution of some key features. Research findings also indicated that resilient students tended to have a stable emotional state, high levels of self-efficacy, low levels of truancy and positive future aspirations. This study has established a research paradigm essentially methodological in nature to bridge the gap between psychological theories and big data in the field of educational psychology. To sum up, our study shed light on the issues of education equity and quality from a global perspective in the times of the COVID-19 pandemic.

Unveiling Carbon Cluster Coating in Graphene CVD on MgO: Combining Machine Learning Force field and DFT Modeling.

In this study, we investigate the behavior of carbon clusters (Cn, where n ranges from 16 to 26) supported on the surface of MgO. We consider the impact of doping with common impurities (such as Si, Mn, Ca, Fe, and Al) that are typically found in ores. Our approach combines density functional theory calculations with machine learning force field molecular dynamics simulations. It is found that the C21 cluster, featuring a core-shell structure composed of three pentagons isolated by three hexagons, demonstrates exceptional stability on the MgO surface and behaves as an "enhanced binding agent" on MgO-doped surfaces. The molecular dynamics trajectories reveal that the stable C21 coating on the MgO surface exhibits less mobility compared to other sizes Cn clusters and the flexible graphene layer on MgO. Furthermore, this stability persists even at temperatures up to 1100K. The analysis of the electron localization function and potential function of Cn on MgO reveals the high localization electron density between the central carbon of the C21 ring and the MgO surface. This work proposes that the C21 island serves as a superstable and less mobile precursor coating on MgO surfaces. This explanation sheds light on the experimental defects observed in graphene products, which can be attributed to the reduced mobility of carbon islands on a substrate that remains frozen and unchanged.

Monitoring and assessment of spatiotemporal soil salinization in the Lake Urmia region

Soil salinization stands as a prominent global environmental challenge, necessitating enhanced assessment methodologies. This study is dedicated to refining soil salinity assessment in the Lake Urmia region of Iran, utilizing multi-year data spanning from 2015 to 2018. To achieve this objective, soil salinity was measured at 915 sampling points during the 2015–2018 timeframe. Simultaneously, remote sensing data were derived from surface reflectance data over the same study period. Four distinct scenarios were considered such as a newly developed spectral index (Scenario I), the newly developed index combined with other salt-based spectral indices from the literature (Scenario II), indirect spectral indices based on vegetation and soil characteristics (Scenario III), and the amalgamation of both direct and indirect spectral indices (Scenario IV). Linear Regression (LR), Support Vector Machine (SVM), and Random Forest (RF) were employed to assess soil salinity. The measured data divided to 75% of the data as the calibration dataset, while the remaining 25% constituted the validation dataset. The findings revealed a correlation between soil salinity and spectral indices from the literature, with a range of -0.53 to 0.51, while the newly developed spectral index exhibited a stronger correlation (r = 0.59). Furthermore, RF yielded superior results when using the newly developed spectral index (Scenario I). Overall, SVM emerged as the most effective model (ME = -9.678, R2 = 0.751, and RPIQ = 1.78) when integrating direct and indirect spectral indices (Scenario IV). This study demonstrates the efficacy of combining machine learning techniques with a blend of newly developed and existing spectral indices from the literature for the monitoring of soil salinity, particularly in arid and semi-arid regions.

Identification of potential biomarkers for lung adenocarcinoma: a study based on bioinformatics analysis combined with validation experiments.

The prognosis for lung adenocarcinoma (LUAD) remains dismal, with a 5-year survival rate of <20%. Therefore, the purpose of this study was to identify potentially reliable biomarkers in LUAD by machine learning combination with Mendelian randomization (MR). TCGA-LUAD, GSE40791, and GSE31210 were employed this study. Key module differential genes were identified through differentially expressed analysis and weighted gene co-expression network analysis (WGCNA). Furthermore, candidate biomarkers were derived from protein-protein interaction network (PPI) and machine learning. Ultimately, biomarkers were confirmed using MR analysis. In addition, immunohistochemistry was used to detect the expression levels of genes that have a causal relationship to LUAD in the LUAD group and the control group. Cell experiments were conducted to validate the effect of screening genes on proliferation, migration, and apoptosis of LUAD cells. The correlation between the screened genes and immune infiltration was determined by CIBERSORT algorithm. In the end, the gene-related drugs were predicted through the Drug-Gene Interaction database. In total, 401 key module differential genes were obtained by intersecting of 5,702 differentially expressed genes (DEGs) and 406 key module genes. Thereafter, GIMAP6, CAV1, PECAM1, and TGFBR2 were identified. Among them, only TGFBR2 had a significant causal relationship with LUAD (p=0.04, b=-0.06), and it is a protective factor for LUAD. Subsequently, sensitivity analyses showed that there were no heterogeneity and horizontal pleiotropy in the univariate MR results, and the results were not overly sensitive to individual SNP loci, further validating the reliability of univariate Mendelian randomization (UVMR) results. However, no causal relationship was found between them by reverse MR analysis. Meanwhile, TGFBR2 expression was decreased in LUAD group through immunohistochemistry. TGFBR2 can inhibit proliferation and migration of lung adenocarcinoma cell line A549 and promote apoptosis of A549 cells. Immune infiltration analysis suggested a potential link between TGFBR2 expression and immune infiltration. Finally, Irinotecan and Hesperetin were predicted through DGIDB database. In this study, TGFBR2 was identified as a biomarker of LUAD, which provided a new idea for the treatment strategy of LUAD and may aid in the development of personalized immunotherapy strategies.

Leveraging Blockchain and Machine Learning to Promote Child Labor-Free Sustainable Development

Child labor has been on the rise in recent years, which necessitates improved identification and reporting mechanisms. Current labor management systems, which are often manual or paper-based, lack traceability, audit, privacy, security, and trust features. This leads to challenges in detecting and reporting violations, particularly in large or remote areas. The persistence of this issue undermines the achievement of Sustainable Development Goals (SDGs) and highlights the important role of Corporate Social Responsibility (CSR) in addressing this challenge. Our paper proposes a solution combining machine learning and blockchain to automate child labor detection and ensure traceable, auditable, private, and secure reporting. Utilizing Decentralized Proxy Re-Encryption (DPRE), Zero-Knowledge Proofs (ZKPs), and oracles on the Ethereum blockchain, with decentralized storage, our approach maintains privacy and transparency. We present a child labor detection model using Mask2Former and ResNet-18 Convolutional Neural Network (CNN) to achieve high accuracy and reliability. The model’s performance is evaluated using various metrics, achieving an accuracy rate of 89.45%, a precision score of 0.906, and a recall score of 0.9332. Additionally, we assess smart contracts for cost-efficiency and security, and discuss the solution’s generalizability, challenges, and practical implications. We make the source code of our solution publicly available on GitHub.