High-performance Machine Learning Research Articles

Characterization and automated classification of sentences in the biomedical literature: a case study for biocuration of gene expression and protein kinase activity.

Biological knowledgebases are essential resources for biomedical researchers, providing ready access to gene function and genomic data. Professional, manual curation of knowledgebases, however, is labor-intensive and thus high-performing machine learning methods that improve biocuration efficiency are needed. Here we report on sentence-level classification to identify biocuration-relevant sentences in the full text of published references for two gene function data types: gene expression and protein kinase activity. We performed a detailed characterization of sentences from references in the WormBase bibliography and used this characterization to define three tasks for classifying sentences as either 1) fully curatable, 2) fully and partially curatable, or 3) all language-related. We evaluated various machine learning (ML) models applied to these tasks and found that GPT and BioBERT achieve the highest average performance, resulting in F1 performance scores ranging from 0.89 to 0.99 depending upon the task. Our findings demonstrate the feasibility of extracting biocuration-relevant sentences from full text. Integrating these models into professional biocuration workflows, such as those used by the Alliance of Genome Resources and the ACKnowledge community curation platform, might well facilitate efficient and accurate annotation of the biomedical literature.

FormulationBCS: A Machine Learning Platform Based on Diverse Molecular Representations for Biopharmaceutical Classification System (BCS) Class Prediction.

The Biopharmaceutics Classification System (BCS) has facilitated biowaivers and played a significant role in enhancing drug regulation and development efficiency. However, the productivity of measuring the key discriminative properties of BCS, solubility and permeability, still requires improvement, limiting high-throughput applications of BCS, which is essential for evaluating drug candidate developability and guiding formulation decisions in the early stages of drug development. In recent years, advancements in machine learning (ML) and molecular characterization have revealed the potential of quantitative structure-performance relationships (QSPR) for rapid and accurate in silico BCS classification. The present study aims to develop a web platform for high-throughput BCS classification based on high-performance ML models. Initially, four data sets of BCS-related molecular properties: log S, log P, log D, and log Papp were curated. Subsequently, 6 ML algorithms or deep learning frameworks were employed to construct models, with diverse molecular representations ranging from one-dimensional molecular fingerprints, descriptors, and molecular graphs to three-dimensional molecular spatial coordinates. By comparing different combinations of molecular representations and learning algorithms, LightGBM exhibited excellent performance in solubility prediction, with an R2 of 0.84; AttentiveFP outperformed others in permeability prediction, with R2 values of 0.96 and 0.76 for log P and log D, respectively; and XGBoost was the most accurate for log Papp prediction, with an R2 of 0.71. When externally validated on a marketed drug BCS category data set, the best-performing models achieved classification accuracies of over 77 and 73% for solubility and permeability, respectively. Finally, the well-trained models were embedded into the first ML-based BCS class prediction web platform (x f), enabling pharmaceutical scientists to quickly determine the BCS category of candidate drugs, which will aid in the high-throughput BCS assessment for candidate drugs during the preformulation stage, thereby promoting reduced risk and enhanced efficiency in drug development and regulation.

Machine Learning and Deep Learning Applications in Disinformation Detection: A Bibliometric Assessment

Fake news is one of the biggest challenging issues in today’s technological world and has a huge impact on the population’s decision-making and way of thinking. Disinformation can be classified as a subdivision of fake news, the main purpose of which is to manipulate and generate confusion among people in order to influence their opinion and obtain certain advantages in multiple domains (politics, economics, etc.). Propaganda, rumors, and conspiracy theories are just a few examples of common disinformation. Therefore, there is an urgent need to understand this phenomenon and offer the scientific community a paper that provides a comprehensive examination of the existing literature, lay the foundation for future research areas, and contribute to the fight against disinformation. The present manuscript provides a detailed bibliometric analysis of the articles oriented towards disinformation detection, involving high-performance machine learning and deep learning algorithms. The dataset has been collected from the popular Web of Science database, through the use of specific keywords such as “disinformation”, “machine learning”, or “deep learning”, followed by a manual check of the papers included in the dataset. The documents were examined using the popular R tool, Biblioshiny 4.2.0; the bibliometric analysis included multiple perspectives and various facets: dataset overview, sources, authors, papers, n-gram analysis, and mixed analysis. The results highlight an increased interest from the scientific community on disinformation topics in the context of machine learning and deep learning, supported by an annual growth rate of 96.1%. The insights gained from the research bring to light surprising details, while the study provides a solid basis for both future research in this area, as well for the development of new strategies addressing this complex issue of disinformation and ensuring a trustworthy and safe online environment.

Optimized Machine Learning Algorithms for Predicting the Punching Shear Resistance of Flat Slabs with Transverse Reinforcement

In the calculation of reinforced concrete (RC) flat slabs with transverse reinforcement, punching shear resistance is one of the most critical factors. It is true that design provisions may be implemented, but they often result in significant biases and deviations from expectations. This study aims to present an optimized machine learning (ML) algorithm for estimating the punching shear resistance. Four machine learning (ML) algorithms (SVR, DT, RF, and XGBoost) with Bayesian optimization (BO) are presented in this paper to provide accurate predictions for flat slabs. The adoptability and optimization of the models are achieved through the analysis of a database of 337 test specimens with nine design parameters. Machine learning (ML) techniques are used to estimate punching shear resistance, which is compared with design provisions and equations relating to critical shear crack theory (CSCT). According to this study, Bayesian optimization is still capable of improving the performance of conventional machine learning algorithms, while the XGBoost-based model offers advanced capabilities. Predictions based on BO-XGBoost are in good agreement with actual values (MAE, RMSE, and R2 are 0.09 MN, 0.14 MN, and 0.92, respectively) in test set. Following a detailed explanation using Shapley additive explanation (SHAP), a high-performance ML approach is used to investigate the predictive results. With the proposed optimized algorithms, it is possible to determine the punching shear resistance of flat slabs with transverse reinforcement during the preliminary stages of the construction.

Identifying Oncology Patients at High Risk for Potentially Preventable Emergency Department Visits Using a Novel Definition.

Patients with cancer visit the emergency department (ED) frequently. While some ED visits are necessary, others may be potentially preventable ED visits (PPEDs). Reducing PPEDs is important to improve quality of care and reduce costs. However, a robust definition and the characteristics of patients at risk remain unclear. This study aimed to describe oncology-related PPEDs and identify characteristics of patients at the highest risk for PPEDs to help target interventions and minimize avoidable ED visits. A retrospective study was conducted using four clinical and administrative databases. All ED visits by oncology patients between April 1, 2019, and April 1, 2021, were identified. A novel definition of PPEDs was explored, specifically visits that resulted in immediate discharge from the ED or admissions <48 hours. Trends in ED use, including PPEDs, were evaluated using descriptive statistics, logistic regression, and machine learning (ML) modeling. During the 2-year period, 6,689 oncology patients visited the ED (N = 13,415 visits). A total of 62.1% of visits were classified as PPEDs. PPEDs were most common among patients with stage I to III breast cancer and those on systemic therapy. Characteristics of patients at high risk for non-PPEDs included stage IV disease with either lung or GI carcinomas and shorter distances to the ED. The highest-performing ML model yielded an AUC of 0.819. Our novel definition of PPEDs appears promising in identifying oncology patients who could avoid the ED with targeted interventions. This work demonstrated that patients with early-stage disease, those with breast cancer, and those on systemic therapy are at the highest risk for PPEDs and may benefit from proactive interventions to avoid the ED. Although our definition requires validation, using ML models for more robust predictive modeling appears promising.

Novel interpretable and robust web-based AI platform for phishing email detection

Phishing emails continue to pose a significant threat, causing financial losses and security breaches. This study addresses limitations in existing research, such as reliance on proprietary datasets and lack of real-world application, by proposing a high-performance machine learning model for email classification. Utilizing a comprehensive and largest available public dataset, the model achieves a f1 score of 0.99 and is designed for deployment within relevant applications. Additionally, Explainable AI (XAI) is integrated to enhance user trust. This research offers a practical and highly accurate solution, contributing to the fight against phishing by empowering users with a real-time web-based application for phishing email detection.

A Machine Learning Model Based on CT Imaging Metrics and Clinical Features to Predict the Risk of Hospital-Acquired Pneumonia After Traumatic Brain Injury.

To develop a validated machine learning (ML) algorithm for predicting the risk of hospital-acquired pneumonia (HAP) in patients with traumatic brain injury (TBI). We employed the Least Absolute Shrinkage and Selection Operator (LASSO) to identify critical features related to pneumonia. Five ML models-Logistic Regression (LR), Extreme Gradient Boosting (XGB), Random Forest (RF), Naive Bayes Classifier (NB), and Support Vector Machine (SVC)-were developed and assessed using the training and validation datasets. The optimal model was selected based on its performance metrics and used to create a dynamic web-based nomogram. In a cohort of 858 TBI patients, the HAP incidence was 41.02%. LR was determined to be the optimal model with superior performance metrics including AUC, accuracy, and F1-score. Key predictive factors included Age, Glasgow Coma Score, Rotterdam Score, D-dimer, and the Systemic Immune Response to Inflammation Index (SIRI). The nomogram developed based on these predictors demonstrated high predictive accuracy, with AUCs of 0.818 and 0.819 for the training and validation datasets, respectively. Decision curve analysis (DCA) and calibration curves validated the model's clinical utility and accuracy. We successfully developed and validated a high-performance ML algorithm to assess the risk of HAP in TBI patients. The dynamic nomogram provides a practical tool for real-time risk assessment, potentially improving clinical outcomes by aiding in early intervention and personalized patient management.

Application of the XGBoost Model with Hyperparameter Tuning for Industry Classification for Job Applicants

The development of technology and changes in job market dynamics have created new challenges in aligning education with industry needs. In this research, the XGBoost model with hyperparameter tuning was applied for industry classification on job applicant data taken from the Kaggle dataset LinkedIn Job Postings in 2023. This dataset consists of 23 attributes with a total of 33,085 job vacancy data points. The experimental results show that both the model without hyperparameter tuning and with GridSearchCV produce the same classification accuracy, which is 0.89 or 89%, with stable precision, recall, and F1-Score values. The best parameters found in this study are colsample_bytree = 1.0, learning_rate = 0.3, max_depth = 6, min_child_weight = 1, n_estimators = 100, and subsample = 1.0. However, cross-validation using k-fold shows a significant increase in accuracy to 0.90, or 90%. This finding confirms that the use of cross-validation can improve the performance estimation of the model more accurately and robustly by utilizing all available data for training and testing. Moreover, the implementation of cross-validation demonstrates the importance of leveraging all data points to enhance model reliability and robustness. Future research can explore alternative hyperparameter tuning methods and apply the model to larger datasets to further validate the generalizability and reliability of the XGBoost model in different application contexts. Thus, this study underscores the significance of rigorous model evaluation techniques in achieving high-performing machine learning models

FlowAtlas: an interactive tool for high-dimensional immunophenotyping analysis bridging FlowJo with computational tools in Julia.

As the dimensionality, throughput and complexity of cytometry data increases, so does the demand for user-friendly, interactive analysis tools that leverage high-performance machine learning frameworks. Here we introduce FlowAtlas: an interactive web application that enables dimensionality reduction of cytometry data without down-sampling and that is compatible with datasets stained with non-identical panels. FlowAtlas bridges the user-friendly environment of FlowJo and computational tools in Julia developed by the scientific machine learning community, eliminating the need for coding and bioinformatics expertise. New population discovery and detection of rare populations in FlowAtlas is intuitive and rapid. We demonstrate the capabilities of FlowAtlas using a human multi-tissue, multi-donor immune cell dataset, highlighting key immunological findings. FlowAtlas is available at https://github.com/gszep/FlowAtlas.jl.git.

Comparison of model feature importance statistics to identify covariates that contribute most to model accuracy in prediction of insomnia.

Sleep is critical to a person's physical and mental health and there is a need to create high performing machine learning models and critically understand how models rank covariates. The study aimed to compare how different model metrics rank the importance of various covariates. A cross-sectional cohort study was conducted retrospectively using the National Health and Nutrition Examination Survey (NHANES), which is publicly available. This study employed univariate logistic models to filter out strong, independent covariates associated with sleep disorder outcome, which were then used in machine-learning models, of which, the most optimal was chosen. The machine-learning model was used to rank model covariates based on gain, cover, and frequency to identify risk factors for sleep disorder and feature importance was evaluated using both univariable and multivariable t-statistics. A correlation matrix was created to determine the similarity of the importance of variables ranked by different model metrics. The XGBoost model had the highest mean AUROC of 0.865 (SD = 0.010) with Accuracy of 0.762 (SD = 0.019), F1 of 0.875 (SD = 0.766), Sensitivity of 0.768 (SD = 0.023), Specificity of 0.782 (SD = 0.025), Positive Predictive Value of 0.806 (SD = 0.025), and Negative Predictive Value of 0.737 (SD = 0.034). The model metrics from the machine learning of gain and cover were strongly positively correlated with one another (r > 0.70). Model metrics from the multivariable model and univariable model were weakly negatively correlated with machine learning model metrics (R between -0.3 and 0). The ranking of important variables associated with sleep disorder in this cohort from the machine learning models were not related to those from regression models.

Efficiency of Using GPUs for Reconstructing the Hydraulic Resistance in River Systems Based on Combination of High Performance Hydrodynamic Simulation and Machine Learning

Efficiency of Using GPUs for Reconstructing the Hydraulic Resistance in River Systems Based on Combination of High Performance Hydrodynamic Simulation and Machine Learning

A Survival Prediction Model of Self-Immolation Based on Machine Learning Techniques.

Self-immolation is one of the violent methods of suicide in developing countries. Predicting the survival of self-immolation patients helps develop therapeutic strategies. Today, machine learning is widely used in diagnosing diseases and predicting the survival of patients. This study aims to provide a model to predict the survival of self-immolation patients using machine learning techniques. A retrospective cross-sectional study was conducted on 445 hospitalized self-immolated patients admitted to a burn hospital between March 2008 and 2019. Python programming language version 3.7 was used for this goal. All possible machine-learning algorithms were used. Gradient Boosting, support vector machine (SVM), random forest, multilayer perceptron (MLP), and k-nearest neighbors algorithm (KNN) were selected as the high-performance machine learning technique for survival prediction, and then they were compared by evaluation metrics such as F1 score, accuracy, sensitivity, specificity, and area under the receiver operating characteristic (ROC) curve (AUC). Based on this comparison, the best model was reported. SVM was the best algorithm. F1 score, accuracy, and AUC for this machine-learning model were 91.8%, 91.9%, and 0.96, respectively. The machine learning model results revealed that surgical procedures, score, length of stay, anatomical region, and gender obtained the most important and had more impact than other factors on patients' survival prediction. In this paper, machine learning algorithms were used to create a model for survival of self-immolation patients. The results of this study can be used as a model for predicting self-immolation patients' survival, better treatment management, and setting up policies and medical decision-making in burn centers.

Accelerating Molecular Vibrational Spectra Simulations with a Physically Informed Deep Learning Model.

In recent years, machine learning (ML) surrogate models have emerged as an indispensable tool to accelerate simulations of physical and chemical processes. However, there is still a lack of ML models that can accurately predict molecular vibrational spectra. Here, we present a highly efficient multitask ML surrogate model termed Vibrational Spectra Neural Network (VSpecNN), to accurately calculate infrared (IR) and Raman spectra based on dipole moments and polarizabilities obtained on-the-fly via ML-enhanced molecular dynamics simulations. The methodology is applied to pyrazine, a prototypical polyatomic chromophore. The VSpecNN-predicted energies are well within the chemical accuracy (1 kcal/mol), and the errors for VSpecNN-predicted forces are only half of those obtained from a popular high-performance ML model. Compared to the ab initio reference, the VSpecNN-predicted frequencies of IR and Raman spectra differ only by less than 5.87 cm-1, and the intensities of IR spectra and the depolarization ratios of Raman spectra are well reproduced. The VSpecNN model developed in this work highlights the importance of constructing highly accurate neural network potentials for predicting molecular vibrational spectra.

Enhancing Edge-Assisted Federated Learning with Asynchronous Aggregation and Cluster Pairing

Federated learning (FL) is widely regarded as highly promising because it enables the collaborative training of high-performance machine learning models among a large number of clients while preserving data privacy by keeping the data local. However, many existing FL frameworks have a two-layered architecture, thus requiring the frequent exchange of large-scale model parameters between clients and remote cloud servers over often unstable networks and resulting in significant communication overhead and latency. To address this issue, we propose to introduce edge servers between the clients and the cloud server to assist in aggregating local models, thus combining asynchronous client–edge model aggregation with synchronous edge–cloud model aggregation. By leveraging the clients’ idle time to accelerate training, the proposed framework can achieve faster convergence and reduce the amount of communication traffic. To make full use of the grouping properties inherent in three-layer FL, we propose a similarity matching strategy between edges and clients, thus improving the effect of asynchronous training. We further propose to introduce model-contrastive learning into the loss function and personalize the clients’ local models to address the potential learning issues resulting from asynchronous local training in order to further improve the convergence speed. Extensive experiments confirm that our method exhibits significant improvements in model accuracy and convergence speed when compared with other state-of-the-art federated learning architectures.

Combining ChatGPT and knowledge graph for explainable machine learning-driven design: a case study

Machine learning has been widely used in design activities, enabling more informed decision-making. However, high-performance machine learning models, often referred to as ‘black-box', result in a lack of explainability regarding predictions. The absence of explainability erodes the trust between designers and these models and hinders human-machine collaboration for desirable design decisions. Explainable AI focuses on creating explanations that are accessible and comprehensible to stakeholders, thereby improving explainability. A recent advancement in the field of explainable AI involves leveraging domain-specific knowledge via knowledge graph. Additionally, the advent of large language models like ChatGPT, acclaimed for their ability to output domain knowledge, perform complex language processing, and support seamless end-user interaction, has the potential to expand the horizons of explainable AI. Inspired by these developments, we propose the novel hybrid method that synergizes ChatGPT and knowledge graph to augment post-hoc explainability in design context. The outcome is the generation of more contextual and meaningful explanations, with the added possibility of further interaction to uncover deeper insights. The effectiveness of the proposed method is illustrated through a case study on customer segmentation.

Assessing multi-source random forest classification and robustness of predictor variables in flooded areas mapping

Flood extent delineation techniques have benefited from the increasing availability of remote sensing imagery, classification techniques and the introduction of geomorphic descriptors derived from Digital Elevation Models (DEM). On the other hand, high-performing Machine Learning (ML) methods have allowed for the development of accurate flood maps by integrating several predictor variables into supervised or unsupervised algorithms. Among others, Random Forest (RF) is a powerful and widely applied ML classifier, providing accurate predictions also with complex datasets and for varying parameters set. In the present study, the effectiveness of this algorithm for mapping flooded areas was evaluated. Various geospatial data sources were integrated, including morphological indicators, such as the Geomorphic Flood Index (GFI), Sentinel-2 bands, multispectral indices, and Sentinel-1 polarizations. The reliability of the predictor variables under different training sample sizes was evaluated and the accuracy of the RF classifier was assessed. Moreover, by exploring the algorithm ability to identify the most important variables, the predictors contributing the most to the classification were identified and their stability for varying training parameters was investigated. To gauge the adaptability and consistency of these features, we applied our analyses to different study areas around the World. The results indicate that certain predictors displayed remarkable stability across different sample sizes and remained robust under various training parameters. However, some variability in the algorithm structure and the features related to the specific complexities of each considered study case was also observed.

Automated machine learning: past, present and future

Automated machine learning (AutoML) is a young research area aiming at making high-performance machine learning techniques accessible to a broad set of users. This is achieved by identifying all design choices in creating a machine-learning model and addressing them automatically to generate performance-optimised models. In this article, we provide an extensive overview of the past and present, as well as future perspectives of AutoML. First, we introduce the concept of AutoML, formally define the problems it aims to solve and describe the three components underlying AutoML approaches: the search space, search strategy and performance evaluation. Next, we discuss hyperparameter optimisation (HPO) techniques commonly used in AutoML systems design, followed by providing an overview of the neural architecture search, a particular case of AutoML for automatically generating deep learning models. We further review and compare available AutoML systems. Finally, we provide a list of open challenges and future research directions. Overall, we offer a comprehensive overview for researchers and practitioners in the area of machine learning and provide a basis for further developments in AutoML.

Machine Learning for Predictive Analysis of Otolaryngology Residency Letters of Recommendation.

Letters of recommendation (LORs) are a highly influential yet subjective and often enigmatic aspect of the residency application process. This study hypothesizes that LORs do contain valuable insights into applicants and can be used to predict outcomes. This pilot study utilizes natural language processing and machine learning (ML) models using LOR text to predict interview invitations for otolaryngology residency applicants. A total of 1642 LORs from the 2022-2023 application cycle were retrospectively retrieved from a single institution. LORs were preprocessed and vectorized using three different techniques to represent the text in a way that an ML model can understand written prose: CountVectorizer (CV), Term Frequency-Inverse Document Frequency (TF-IDF), and Word2Vec (WV). Then, the LORs were trained and tested on five ML models: Logistic Regression (LR), Naive Bayes (NB), Decision Tree (DT), Random Forest (RF), and Support Vector Machine (SVM). Of the 337 applicants, 67 were interviewed and 270 were not interviewed. In total, 1642 LORs (26.7% interviewed) were analyzed. The two best-performing ML models in predicting interview invitations were the TF-IDF vectorized DT and CV vectorized DT models. This preliminary study revealed that ML models and vectorization combinations can provide better-than-chance predictions for interview invitations for otolaryngology residency applicants. The high-performing ML models were able to classify meaningful information from the LORs to predict applicant interview invitation. The potential of an automated process to help predict an applicant's likelihood of obtaining an interview invitation could be a valuable tool for training programs in the future. N/A Laryngoscope, 134:4016-4022, 2024.

Can we predict pathology without surgery? Weighing the added value of multiparametric MRI and whole prostate radiomics in integrative machine learning models.

To test the ability of high-performance machine learning (ML) models employing clinical, radiological, and radiomic variables to improve non-invasive prediction of the pathological status of prostate cancer (PCa) in a large, single-institution cohort. Patients who underwent multiparametric MRI and prostatectomy in our institution in 2015-2018 were considered; a total of 949 patients were included. Gradient-boosted decision tree models were separately trained using clinical features alone and in combination with radiological reporting and/or prostate radiomic features to predict pathological T, pathological N, ISUP score, and their change from preclinical assessment. Model behavior was analyzed in terms of performance, feature importance, Shapley additive explanation (SHAP) values, and mean absolute error (MAE). The best model was compared against a naïve model mimicking clinical workflow. The model including all variables was the best performing (AUC values ranging from 0.73 to 0.96 for the six endpoints). Radiomic features brought a small yet measurable boost in performance, with the SHAP values indicating that their contribution can be critical to successful prediction of endpoints for individual patients. MAEs were lower for low-risk patients, suggesting that the models find them easier to classify. The best model outperformed (p ≤ 0.0001) clinical baseline, resulting in significantly fewer false negative predictions and overall was less prone to under-staging. Our results highlight the potential benefit of integrative ML models for pathological status prediction in PCa. Additional studies regarding clinical integration of such models can provide valuable information for personalizing therapy offering a tool to improve non-invasive prediction of pathological status. The best machine learning model was less prone to under-staging of the disease. The improved accuracy of our pathological prediction models could constitute an asset to the clinical workflow by providing clinicians with accurate pathological predictions prior to treatment. • Currently, the most common strategies for pre-surgical stratification of prostate cancer (PCa) patients have shown to have suboptimal performances. • The addition of radiological features to the clinical features gave a considerable boost in model performance. Our best model outperforms the naïve model, avoiding under-staging and resulting in a critical advantage in the clinic. •Machine learning models incorporating clinical, radiological, and radiomics features significantly improved accuracy of pathological prediction in prostate cancer, possibly constituting an asset to the clinical workflow.

Privacy-preserving culvert predictive models: A federated learning approach

In recent years, transportation agencies have shown a growing interest in the adoption of machine learning (ML) models as a means to improve the efficiency of their infrastructure asset management practices. Their limited inventory can, however, hinder the development of robust ML models. Moreover, transportation agencies may be reluctant to share their raw inventory data with others. To address these challenges, we employed a new ML paradigm, federated learning (FL), and demonstrated its potential in the case study of Utah culvert management. The Utah Department of Transportation (UDOT) only owned a limited culvert inventory, but we were able to obtain additional data from five other state DOTs’ inventories. Rather than employing a conventional centralized ML method, we opted for FL. This approach ensured that while UDOT did not directly access raw data from five other DOTs, it could still benefit from their combined insights. The FL-based model with 80.4 % accuracy showed a promising performance between the two centralized learning models developed based on the consolidated and the Utah-only datasets. The federated approach demonstrates that it is possible to develop high-performance ML models without compromising data confidentiality by centralizing data. This pioneering approach sets a precedent for other transportation agencies, suggesting they can navigate data scarcity and privacy concerns while reaping the benefits of collective knowledge.