Train Machine Learning Models Research Articles

Robust quantum federated learning with noise

Abstract Federated learning is a decentralized machine learning approach where multiple participants collaboratively train machine learning models. With the development of quantum computing, there has been significant potential in the integration of quantum computing and federated learning. However, existing research has demonstrated that, similar to classical federated learning models, quantum federated learning models also face various security threats and privacy leakage issues. This paper proposes a quantum federated learning model based on quantum noise. Adding quantum noise to the model not only addresses privacy leakage, but also enhances the model robustness, effectively resists adversarial attacks. Specifically, extensive numerical simulations are conducted using various datasets to evaluate the effectiveness of the proposed method. The results reveal a more pronounced variation in robust training in high-dimensional datasets compared to low-dimensional datasets. Furthermore, the impact of noise intensity on model robustness is explored. Experimental demonstrate that a small amount of quantum noise does not have a significant impact on accuracy, and as the noise increases, the robustness of the model also improves. Finally, three different types of quantum noise were used for robustness testing in the paper to analyze the impact of quantum noise on the robustness of quantum machine learning models. The abundant experimental results have verified that the noise can improve the security of distributed quantum machine learning.

AntiMicro.ai: An Artificial Intelligence powered web application for predicting antibacterial/antifungal susceptibility and constructing personalized machine learning models

Background To train and develop machine learning models on the Pfizer antibacterial and antifungal datasets with subsequent deployment to an interactive Web Application. Methods We utilized R version 4.3.1 to perform descriptive analysis to obtain features/predictors. Python 3.10 libraries NumPy, Pandas, Scikit learn, Pycaret were used to train machine learning models. All these models were scored using area under the curve, recall, precision, F1, Kappa and the Mathews correlation coefficient. The best performing model was then deployed into a web application built on Streamlit after which deployment was done using GitHub and Streamlit cloud. A prototype android app was also developed using GoNative. Results The exploratory data analysis, S Aureus (17.2%) was the most common species however in the sub group analysis of the isolates with genotypic values, K Pneumoniae(48.2%) and E Coli (36.9%) were dominant. Amongst the fungi, Candida albicans (38.3%) and Candida glabrata (15.5%) were dominant. Feature selection was done using Shapley additive explanation plots. Using Extreme Gradient Boosting (XGBoost), we were able to achieve 99% and 97.8% AUC in the prediction of antibacterial and antifungal susceptibility respectively with minimal overfitting. Conclusions Decision tree based methods are a viable options of predicting antibacterial and antifungal drug resistance. When presented in visually appealing modes like web applications and android apps, it can be a useful guide to clinicians on initial treatment while awaiting definitive phenotypic testing. It can also be a surveillance tool that can craft antimicrobial resistance strategies.

SchNetPack Hyperparameter Optimization for a More Reliable Top Docking Scores Prediction.

Options to improve the extrapolation power of the neural network designed using the SchNetPack package with respect to top docking scores prediction are presented. It is shown that hyperparameter tuning of the atomistic model representation (in the schnetpack.representation) improves the prediction of the top scoring compounds, which have characteristically a low incidence in randomized data sets for training of machine learning models. The prediction robustness is evaluated according to the mean square error (MSE) and the entropy of the average loss landscape decrease. Admittedly, the improvement of the top scoring compounds' prediction accuracy comes with the penalty of worsening the overall prediction power. It is revealed that the most impactful hyperparameter is the cutoff (5 Å is reported as the optimal choice). Other parameters (e.g., number of radial basis functions, number of interaction layers of the neural network, feature vector size or its batch size) are found to not affect the prediction robustness of the top scoring compounds in any comparable way relative to the cutoff. The MSE of the best docking score prediction (below -13 kcal/mol) improves from ca. 3.5 to 0.9 kcal/mol, while the prediction of less potent compounds (-13 to -11 kcal/mol) shows a lesser improvement, i.e., a decrease of MSE from 1.6 to 1.3 kcal/mol. Additionally, oversampling and undersampling of the training set with respect to the top scoring compounds' abundance is presented. The results indicate that the cutoff choice performs better than over- or undersampling of the training set, with undersampling performing better than oversampling.

Dynamic Playlist Generation using Machine Learning

This study investigates the application of machine learning algorithms for dynamic playlist generation in the context of music recommendation systems. The traditional method of creating playlists frequently depends on immutable standards, like artist or genre, which might not adequately represent the unique and dynamic character of personal preferences. On the other hand, our suggested system makes use of machine learning methods to examine user behavior, preferences, and contextual elements in order to create playlists that are dynamically created and customized to each user’s individual preferences. In order to train machine learning models, the study collects user interaction data, such as listening history, skip patterns, and user feedback. To extract patterns and relationships from the data, a variety of algorithms are used, including hybrid models, content-based filtering, and collaborative filtering. After that, the models are incorporated into a dynamic playlist creation system that can eventually adjust to changing user preferences.Our test findings show how well the suggested strategy works to improve user experience by offering more interesting and customized playlists. With its ability to adjust to changing user preferences and contextual cues, the dynamic playlist generation system provides a smooth and pleasurable music discovery experience. We also talk about possible enhancements, implementation difficulties, and deployment considerations in the real world.This work adds to the ongoing efforts to improve music recommendation systems by demonstrating how machine learning can be used to develop more responsive and intelligent playlist generation systems. The results highlight how crucial customized experiences are in the constantly changing world of digital music consumption. Index Terms—hybrid mode, content-based filtering, intelligent playlist, music

ECG-image-kit: a synthetic image generation toolbox to facilitate deep learning-based electrocardiogram digitization

Abstract Objective: Cardiovascular diseases are a major cause of mortality globally, and electrocardiograms (ECGs) are crucial for diagnosing them. Traditionally, ECGs are stored in printed formats. However, these printouts, even when scanned, are incompatible with advanced ECG diagnosis software that require time-series data. Digitizing ECG images is vital for training machine learning models in ECG diagnosis, leveraging the extensive global archives collected over decades. Deep learning models for image processing are promising in this regard, although the lack of clinical ECG archives with reference time-series data is challenging. Data augmentation techniques using realistic generative data models provide a solution.

Approach: We introduce ECG-Image-Kit, an open-source toolbox for generating synthetic multi-lead ECG images with realistic artifacts from time-series data, aimed at automating the conversion of scanned ECG images to ECG data points. The tool synthesizes ECG images from real time-series data, applying distortions like text artifacts, wrinkles, and creases on a standard ECG paper background.

Main results: As a case study, we used ECG-Image-Kit to create a dataset of 21,801 ECG images from the PhysioNet QT database. We developed and trained a combination of a traditional computer vision and deep neural network model on this dataset to convert synthetic images into time-series data for evaluation. We assessed digitization quality by calculating the signal-to-noise ratio (SNR) and compared clinical parameters like QRS width, RR, and QT intervals recovered from this pipeline, with the ground truth extracted from ECG time-series. The results show that this deep learning pipeline accurately digitizes paper ECGs, maintaining clinical parameters, and highlights a generative approach to digitization.

Significance: The toolbox has broad applications, including model development for ECG image digitization and classification. The toolbox currently supports data augmentation for the 2024 PhysioNet Challenge, focusing on digitizing and classifying paper ECG images.

Machine Learning Library in C++

This project aims to develop a comprehensive and versatile machine learning library in C++ tailored to address the diverse needs of developers and researchers in the field. The library encompasses a robust set of core machine learning algorithms, encompassing supervised, unsupervised, and reinforcement learning techniques. Additionally, it incorporates essential data preprocessing tools to streamline data manipulation and feature engineering tasks, along with model evaluation capabilities crucial for assessing algorithm performance. The library's primary focus is on providing a rich suite of machine learning algorithms, This empowers users to effectively prepare data for training machine learning models. Additionally, the library provides tools for data splitting, and model evaluation to ensure reliable and robust model performance assessment.

SYN-GAN: A robust intrusion detection system using GAN-based synthetic data for IoT security

As technological communication progresses, diverse datasets are exchanged across distributed environments using the Internet of Things (IoT). However, the IoT environment is vulnerable to attacking and breaching data privacy or making a robust system worse by providing attack data. To address potential risks of attacks, researchers have been conducting experiments on network intrusion detection systems (NIDS) to mitigate threats effectively. The issue of data imbalance and associated data collection costs persists, hindering the ability of machine learning (ML) models to learn malicious behaviour effectively and consequently impacting the accuracy of network threat detection. Addressing these issues, our study explores the potential of using 100% synthetic data generated via Generative Adversarial Networks (GAN) for training ML models in Network Intrusion Detection Systems (NIDS). This approach reduces the dependency on real-world data significantly, paving the way for a more flexible and ethically convenient model-building process. For the UNSW-NB15 dataset, we achieved an accuracy of 90%, a precision of 91%, a recall of 90%, and an F1 score of 89%. For the NSL-KDD dataset, our results showed an accuracy of 84%, a precision of 85%, a recall of 84%, and an F1 score of 84%. For the BoT-IoT dataset, we attained perfect scores of 100% across all metrics. These outcomes indicate that the values obtained from our analysis demonstrate high performance, yielding comparative or superior results to previous studies. Therefore, our study successfully replicates real-world network intrusion detection data, showing new opportunities for the use of generative data in cyber security.

Multimodal emotion classification using machine learning in immersive and non-immersive virtual reality

Affective computing has been widely used to detect and recognize emotional states. The main goal of this study was to detect emotional states using machine learning algorithms automatically. The experimental procedure involved eliciting emotional states using film clips in an immersive and non-immersive virtual reality setup. The participants’ physiological signals were recorded and analyzed to train machine learning models to recognize users’ emotional states. Furthermore, two subjective ratings emotional scales were provided to rate each emotional film clip. Results showed no significant differences between presenting the stimuli in the two degrees of immersion. Regarding emotion classification, it emerged that for both physiological signals and subjective ratings, user-dependent models have a better performance when compared to user-independent models. We obtained an average accuracy of 69.29 ± 11.41% and 71.00 ± 7.95% for the subjective ratings and physiological signals, respectively. On the other hand, using user-independent models, the accuracy we obtained was 54.0 ± 17.2% and 24.9 ± 4.0%, respectively. We interpreted these data as the result of high inter-subject variability among participants, suggesting the need for user-dependent classification models. In future works, we intend to develop new classification algorithms and transfer them to real-time implementation. This will make it possible to adapt to a virtual reality environment in real-time, according to the user’s emotional state.

Audio Recognition of the Percussion Sounds Generated by a 3D Auto-Drum Machine System via Machine Learning

Audio Recognition of the Percussion Sounds Generated by a 3D Auto-Drum Machine System via Machine Learning

Approach to Detecting Malicious Bots in the Vkontakte Social Network and Assessing Their Parameters

The emergence of new varieties of bots in social networks and the improvement of their capabilities to imitate the natural behavior of real users represent a significant problem in the field of protection of social networks and online communities. This paper proposes a new approach to detecting and assessing the parameters of bots within the social network «VKontakte». The basis of the proposed approach is the creation of datasets using the method of «controlled purchase» of bots, which allows one to assess bots’ characteristics such as price, quality, and speed of action of bots, and using the Turing Test to assess how much users trust bots. In combination with traditional machine learning methods and features extracted from interaction graphs, text messages, and statistical distributions, it becomes possible to not only detect bots accurately but also predict their characteristics. This paper demonstrates that the trained machine learning model, based on the proposed approach, is robust to imbalanced data and can identify most types of bots as it has only a minor correlation with their main characteristics. The proposed approach can be used within the choice of countermeasures for the protection of social networks and for historical analysis, which allows not only to confirm the presence of bots but also to characterize the specifics of the attack.

Deep neural network based distortion parameter estimation for blind quality measurement of stereoscopic images

Stereoscopic/3D image quality measurement (SIQM) has emerged as an active and important research branch in image processing/computer vision field. Existing methods for blind/no-reference SIQM often train machine-learning models on degraded stereoscopic images for which human subjective quality ratings have been obtained, and they are thus constrained by the fact that only a limited number of 3D image quality datasets currently exist. Although methods have been proposed to overcome this restriction by predicting distortion parameters rather than quality scores, the approach is still limited to the time-consuming, hand-crafted features extracted to train the corresponding classification/regression models as well as the rather complicated binocular fusion/rivalry models used to predict the cyclopean view. In this paper, we explore the use of deep learning to predict distortion parameters, giving rise to a more efficient opinion-unaware SIQM technique. Specifically, a deep fusion-and-excitation network which takes into account the multiple-distortion interactions is proposed to perform distortion parameter estimation, thus avoiding hand-crafted features by using convolution layers while simultaneously accelerating the algorithm by using the GPU. Moreover, we measure distortion parameter values of the cyclopean view by using support vector regression models which are trained on the data obtained from a newly-designed subjective test. In this way, the potential errors in computing the disparity map and cyclopean view can be prevented, leading to a more rapid and precise 3D-vision distortion parameter estimation. Experimental results tested on various 3D image quality datasets demonstrate that our proposed method, in most cases, offers improved predictive performance over existing state-of-the-art methods.

Machine learning to predict notes for chart review in the oncology setting: a proof of concept strategy for improving clinician note-writing.

Leverage electronic health record (EHR) audit logs to develop a machine learning (ML) model that predicts which notes a clinician wants to review when seeing oncology patients. We trained logistic regression models using note metadata and a Term Frequency Inverse Document Frequency (TF-IDF) text representation. We evaluated performance with precision, recall, F1, AUC, and a clinical qualitative assessment. The metadata only model achieved an AUC 0.930 and the metadata and TF-IDF model an AUC 0.937. Qualitative assessment revealed a need for better text representation and to further customize predictions for the user. Our model effectively surfaces the top 10 notes a clinician wants to review when seeing an oncology patient. Further studies can characterize different types of clinician users and better tailor the task for different care settings. EHR audit logs can provide important relevance data for training ML models that assist with note-writing in the oncology setting.

Analysis of machine learning prediction reliability based on sampling distance evaluation with feature decorrelation

Despite successful use in a wide variety of disciplines for data analysis and prediction, machine learning (ML) methods suffer from a lack of understanding of the reliability of predictions due to the lack of transparency and black-box nature of ML models. In materials science and other fields, typical ML model results include a significant number of low-quality predictions. This problem is known to be particularly acute for target systems which differ significantly from the data used for ML model training. However, to date, a general method for uncertainty quantification (UQ) of ML predictions has not been available. Focusing on the intuitive and computationally efficient similarity-based UQ, we show that a simple metric based on Euclidean feature space distance and sampling density together with the decorrelation of the features using Gram–Schmidt orthogonalization allows effective separation of the accurately predicted data points from data points with poor prediction accuracy. To demonstrate the generality of the method, we apply it to support vector regression models for various small data sets in materials science and other fields. We also show that this metric is a more effective UQ tool than the standard approach of using the average distance of k nearest neighbors (k = 1–10) in features space for similarity evaluation. Our method is computationally simple, can be used with any ML learning method and enables analysis of the sources of the ML prediction errors. Therefore, it is suitable for use as a standard technique for the estimation of ML prediction reliability for small data sets and as a tool for data set design.

Abstract PO1-01-11: Development and Validation of a Breast Cancer Recurrence Model Demonstrates Accurate Identification of Patients with Favorable Long-Term Outcomes

Abstract Background: The Oncotype DX (ODX) test is a 21-gene expression assay widely used for the prediction of risk recurrence in early-stage breast cancer, but it may be possible to identify patients who can forgo testing using only clinicopathologic variables. In 2018, the National Cancer Database (NCDB) began reporting quantitative histologic parameters for estrogen receptor (ER), progesterone receptor (PR), and Ki-67 expression in breast cancer patients. Inclusion of these variables may improve the development of nationally applicable models to predict ODX results using clinicopathologic variables alone. Methods: Using a cohort of patients from the NCDB diagnosed from 2018–2020 with hormone receptor (HR)-positive, HER2-negative, Stage I-III breast cancer, we trained machine learning models to predict high-risk (26-100) ODX score. A subset comprising 80% of patients was used for model training, while the remaining data were set aside for internal validation. An external validation cohort was selected from the University of Chicago Medical Center (UCMC), including patients diagnosed from 2009–2021. Feature selection, model architecture selection, and hyperparameter tuning were performed using 10-fold cross-validation within the NCDB training set. We compared a model with quantitative ER, PR, and Ki-67; a model with only quantitative ER and PR, and a model without quantitative immunohistochemistry – to best reflect the likely data available in a variety of practice patterns. The primary endpoint was the area under the receiver operating characteristic curve (AUROC) for prediction of high-risk ODX results in the UCMC validation cohort. Models were also evaluated as rule-out tests to identify low-risk patients who did not require further ODX testing, using a high (90%) sensitivity threshold, fit in the NCDB training dataset. Results: We identified 53,346 patients from the NCDB cohort meeting the inclusion criteria; 7% had a high risk ODX score, with a median follow-up time of 28 months. The UCMC validation cohort included 896 patients, and was more diverse, with 30% non-Hispanic Black patients (versus 8% in NCDB), more high-risk patients (18% with high ODX), and a longer median follow-up time of 55 months. In the NCDB validation cohort, models incorporating quantitative ER/PR (AUROC 0.78, 95% CI 0.77–0.80) and quantitative ER/PR/Ki-67 (AUROC 0.81, 95% CI 0.80–0.83) both performed better than the non-quantitative model (AUROC 0.70, 95% CI 0.68–0.72). These results were preserved in the external UCMC cohort, where the ER/PR model (AUROC 0.86, 95% CI 0.80–0.92, p = 0.032) and the ER/PR/Ki-67 model (AUROC 0.87, 95% CI 0.81–0.93) outperformed the non-quantitative model (AUROC 0.80, 95% CI 0.73–0.87, p = 0.009). The high sensitivity rule-out threshold of the ER/PR model predicted that 30% of patients in the UCMC cohort would be low ODX, and the ER/PR/Ki-67 model predicted 44% as low risk – negative predictive value was over 96% for prediction of high ODX. Of the patients predicted to be low risk by the quantitative models, none had a documented high ODX score, and recurrence was < 3% at 5 years. The hazard ratio for recurrence free interval, adjusted for age and comorbidity score, of patients predicted to be high risk by this threshold was 2.96 (95% CI 1.02–8.58) for the ER/PR model and 3.84 (95% CI 1.48–9.97) for the ER/PR/Ki-67 model. Conclusions: We present externally validated and nationally applicable models that identify approximately half of HR-positive/HER2-negative breast cancer patients who are unlikely to have high ODX results using widely available quantitative clinicopathologic variables. Patients identified as low risk by these models have excellent long-term outcomes and may be able to forgo adjuvant chemotherapy without further genomic testing. Citation Format: Asim Dhungana, Augustin Vannier, Fangyuan Zhao, Jincong Freeman, Poornima Saha, Megan Sullivan, Katharine Yao, Elbio Flores, Olufunmilayo Olopade, Dezheng Huo, Alexander Pearson, Frederick Howard. Development and Validation of a Breast Cancer Recurrence Model Demonstrates Accurate Identification of Patients with Favorable Long-Term Outcomes [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO1-01-11.

Reactivities of acrylamide warheads toward cysteine targets: a QM/ML approach to covalent inhibitor design.

Covalent inhibition offers many advantages over non-covalent inhibition, but covalent warhead reactivity must be carefully balanced to maintain potency while avoiding unwanted side effects. While warhead reactivities are commonly measured with assays, a computational model to predict warhead reactivities could be useful for several aspects of the covalent inhibitor design process. Studies have shown correlations between covalent warhead reactivities and quantum mechanic (QM) properties that describe important aspects of the covalent reaction mechanism. However, the models from these studies are often linear regression equations and can have limitations associated with their usage. Applications of machine learning (ML) models to predict covalent warhead reactivities with QM descriptors are not extensively seen in the literature. This study uses QM descriptors, calculated at different levels of theory, to train ML models to predict reactivities of covalent acrylamide warheads. The QM/ML models are compared with linear regression models built upon the same QM descriptors and with ML models trained on structure-based features like Morgan fingerprints and RDKit descriptors. Experiments show that the QM/ML models outperform the linear regression models and the structure-based ML models, and literature test sets demonstrate the power of the QM/ML models to predict reactivities of unseen acrylamide warhead scaffolds. Ultimately, these QM/ML models are effective, computationally feasible tools that can expedite the design of new covalent inhibitors.

Communication cost-aware client selection in online federated learning: A Lyapunov approach

The proliferation of intelligence services brings data breaches and privacy infringement concerns. To preserve data privacy when training machine learning models, the federated learning (FL) paradigm emerges. Most existing works assume that training data on FL clients are static during the entire learning process. Nevertheless, various real-time intelligent services call for timely processing of continuously generated data, which fosters the advent of online federated learning (OFL). Currently, how to reconcile model utility and communication cost in OFL is still an open problem. To address this challenge, we leverage the Lyapunov optimization framework to devise a novel Low Cost Client Selection (LCCS) algorithm for OFL, which can judiciously select participating clients to maximize model utility with a low communication cost. Specifically, we design the objective as the sum of a penalty function and a Lyapunov drift function to take both gradient-based client valuation and communication cost into account. By minimizing the objective, we further design the LCCS algorithm, which is lightweight for execution on clients. At last, we conduct extensive experiments with traces generated from public datasets. The experimental results demonstrate that LCCS achieves the highest model utility with a fixed communication cost in comparison with the state-of-the-art baselines.

HOG-CNN based evaluation of cement integrity using 2D dispersion curves from an experimental through tubing logging setup

In oil wells, adequate cementation is essential to guarantee that the well has support for the casing and is able to isolate the well from groundwater. Operations known as Plugging & Abandonment (P&A) must carry out inspections in the cement layer of the well to ensure that the bond quality of the cement with the formation and the hydraulic isolation is adequate before well sealing. The interpretation of data logs during the decommissioning processes is made exclusively for a petrophysicists team, which makes the process susceptible to human errors. Furthermore, it requires a lot of analysis time, increasing the search for alternative solutions that help reduce errors in the interpretation of data logs. So far, the use of machine learning has proven to be a strong candidate in this search. This paper deals with frequency domain images (dispersion curves and Cepstrum) that are acquired from acoustic pressure signals obtained in the time domain and along the vertical axis of an experimental setup. These images are used to train machine learning models to test and compare the accuracies obtained. An adjustment of the hyperparameters was carried out to find those that best describe the simulated experimental characteristics. Furthermore, the convolution neural network (CNN) model was combined with the oriented gradient histogram technique (HOG) to verify its efficiency by analyzing experimental data. Two different databases were used in this work, one with 130 samples relating to the average results of the four hydrophones used in the experiment and one with 520 samples relating to all results. The results found demonstrated that when using the dispersion curves as input with the CNN the accuracy hits 100%. Other combinations of models were also tested hitting similar results. This article contributes to the current research scenario as it presents an innovative combination of machine learning techniques applied to the oil and gas field (HOG-CNN) with the proposal to compare different types of databases, such as using Cepstrum instead of dispersion curves and image processing.

Elastic Federated Learning with Kubernetes Vertical Pod Autoscaler for edge computing

Federated Learning (FL) is an emerging paradigm for training machine learning models across decentralized edge devices, ensuring data privacy and reducing computational tasks on the central cloud. However, the dynamic and resource-constrained nature of edge environments raises significant challenges in deploying FL applications efficiently. This paper introduces an Elastic Federated Learning framework that leverages Kubernetes Vertical Pod Autoscaler to improve the performance of FL applications. The framework enables dynamic adjustment of computational resources, facilitating effective resource utilization for model training and allowing quicker attainment of the targeted accuracy levels for the trained models. Our experiments demonstrate that the proposed framework significantly enhances FL efficiency, maintains model training progress during resource adjustments, and effectively addresses the resource allocation challenges in edge environments. This work advances the capabilities of FL in edge computing, opening doors to more efficient and scalable AI applications while preserving data privacy.

Deep Learning of a 200-member Ensemble with a Limited Historical Training to Improve the Prediction of Extreme Precipitation Events

Abstract This study introduces a deep learning (DL) scheme to generate reliable and skillful probabilistic quantitative precipitation forecasts (PQPFs) in a postprocessing framework. Enhanced machine learning model architecture and training mechanisms are proposed to improve reliability and skill of PQPFs while permitting computationally efficient model fitting using a short training dataset. The methodology is applied to postprocessing of 24h accumulated PQPFs from an ensemble forecast system recently introduced by the Center for Western Weather and Water Extremes (CW3E) and for lead times from one to six days. The ensemble system was designed based on a high-resolution version of the Weather Research and Forecasting (WRF) model, named West-WRF, to produce a 200-member ensemble in near real-time (NRT) over the Western U.S. during the boreal cool seasons to support forecastinformed reservoir operations (FIRO) and studies of prediction of heavy-to-extreme events. Postprocessed PQPFs are compared with those from the raw West-WRF ensemble, the operational Global Ensemble Forecast System (GEFSv12) as well as the ensemble from the European Centre for Medium-Range Weather Forecasts (ECMWF). As an additional baseline, we provide PQPFs verification metrics from a recently developed neural network postprocessing scheme proposed by Ghazvinian et al. (2022). Results demonstrate that skill of postprocessed forecasts significantly outperform PQPFs and deterministic forecasts from raw ensembles and the recently developed algorithm. The resulting PQPFs broadly improve upon the reliability and skill of baselines in predicting heavy to extreme precipitation (e.g., > 75 mm) across all lead times while maintaining the spatial structure of the high-resolution raw ensemble.

QuanDB: a quantum chemical property database towards enhancing 3D molecular representation learning.

Previous studies have shown that the three-dimensional (3D) geometric and electronic structure of molecules play a crucial role in determining their key properties and intermolecular interactions. Therefore, it is necessary to establish a quantum chemical (QC) property database containing the most stable 3D geometric conformations and electronic structures of molecules. In this study, a high-quality QC property database, called QuanDB, was developed, which included structurally diverse molecular entities and featured a user-friendly interface. Currently, QuanDB contains 154,610 compounds sourced from public databases and scientific literature, with 10,125 scaffolds. The elemental composition comprises nine elements: H, C, O, N, P, S, F, Cl, and Br. For each molecule, QuanDB provides 53 global and 5 local QC properties and the most stable 3D conformation. These properties are divided into three categories: geometric structure, electronic structure, and thermodynamics. Geometric structure optimization and single point energy calculation at the theoretical level of B3LYP-D3(BJ)/6-311G(d)/SMD/water and B3LYP-D3(BJ)/def2-TZVP/SMD/water, respectively, were applied to ensure highly accurate calculations of QC properties, with the computational cost exceeding 107 core-hours. QuanDB provides high-value geometric and electronic structure information for use in molecular representation models, which are critical for machine-learning-based molecular design, thereby contributing to a comprehensive description of the chemical compound space. As a new high-quality dataset for QC properties, QuanDB is expected to become a benchmark tool for the training and optimization of machine learning models, thus further advancing the development of novel drugs and materials. QuanDB is freely available, without registration, at https://quandb.cmdrg.com/ .