Supervised Machine Learning Methods Research Articles

Knowledge-transfer-enabled porosity prediction for new part geometry in laser metal deposition

Data-driven porosity prediction in Laser Metal Deposition (LMD) is mainly done with supervised machine learning methods. These methods require labeled thermal signatures for model training, with the “labels” being post-process evaluations of porosity. In practice, acquiring porosity records for newly printed parts is expensive and time-consuming; matching thermal signatures from the printing process with the porosity records is subject to data registration errors. To enable convenient porosity prediction for new part geometry, this study proposes a “knowledge transfer” method to transfer prior statistical knowledge about printed parts to new printing processes. The prior knowledge is leveraged to evaluate the statistical property of new thermal signatures and assign them labels. Supervised machine learning methods can be readily trained with labeled data. The effort for post-process porosity inspection is therefore saved, and the efficiency of data-driven porosity prediction is significantly improved. The proposed method is validated with datasets from an LMD machine, specifically an OPTOMEC LENS 750 system. The statistical inference knowledge about a Ti-6Al-4V thin wall is transferred to two different Ti-6Al-4V cylinders, respectively, to label their thermal signatures and train Convolutional Neural Networks (CNNs) for in-situ porosity prediction. The case study results demonstrate the effectiveness of the proposed method.

Assessing innovation capabilities of manufacturing companies by combination of unsupervised and supervised machine learning approaches

This study suggests the use of unsupervised and supervised machine learning algorithms to categorize companies according to their innovation capabilities. Companies are categorized into three groups: good, satisfactory, and unsatisfactory, in order to create a thorough and reliable assessment procedure. In this study, unsupervised and supervised machine learning methods are used to solve an innovation capability evaluation problem. Data is provided via a survey which is performed in manufacturing industry in Turkiye Firstly, dimensions of innovation capability were determined Principal Component Analysis (PCA). Then data labels were determined by k-means clustering algorithm which is an unsupervised learning technique. A model is first trained using data provided via questionnaire survey, and it is then tested using fresh, unused data. The model is trained using classification algorithms including KNN, GaussianNB, RandomForest, Gradient Boosting, AdaBoost, DesisionTree, XGBOOST and LightGBMC, MLPC, and SVMC and its performance is evaluated against test data. Each classification techniques are evaluated using the performance metrics. With the highest accuracy rate of 93% and lowest MAE, MSE and RMSE values, The LightGBMC and SVMC methods were found the most efficient supervised learning method for innovation capability evaluation.

Serological evaluation of risk factors for exposure to malaria in a pre-elimination setting in Malaysian Borneo

Malaysia has reported no indigenous cases of P. falciparum and P. vivax for over 3 years. When transmission reaches such low levels, it is important to understand the individuals and locations where exposure risks are high, as they may be at greater risk in the case of a resurgence of transmission. Serology is a useful tool in low transmission settings, providing insight into exposure over longer durations than PCR or RDT. We ran blood samples from a 2015 population-based survey in northern Sabah, Malaysian Borneo on a multiplex bead assay. Using supervised machine learning methods, we characterised recent and historic exposure to Plasmodium falciparum and P. vivax and found recent exposure to P. falciparum to be very low, with exposure to both species increasing with age. We performed a risk-factor assessment on environmental, behavioural, demographic and household factors, and identified forest activity and longer travel times to healthcare as common risk-factors for exposure to P. falciparum and P. vivax. In addition, we used remote-sensing derived data and geostatistical models to assess environmental and spatial associations with exposure. We created predictive maps of exposure to recent P. falciparum in the study area and showed 3 clear foci of exposure. This study provides useful insight into the environmental, spatial and demographic risk factors for P. falciparum and P. vivax at a period of low transmission in Malaysian Borneo. The findings would be valuable in the case of resurgence of human malarias in the region.

Seismic Characterization of the Blue Mountain Geothermal Field

Subsurface characterization is crucial for geothermal energy exploration and production. Yet hydrothermal reservoirs usually reside in highly fractured and faulted zones where accurate characterization is very challenging because of low signal-to-noise ratios of land seismic data and lack of coherent reflection signals. We perform an active-source seismic characterization for the Blue Mountain geothermal field in Nevada using active seismic data to reveal the elastic medium property complexity and fault distribution at this field. We first employ an unsupervised machine learning method to attenuate groundroll and near-surface guided-wave noise and enhance coherent reflection and scattering signals from noisy seismic data. We then build a smooth initial P-wave velocity model based on an existing magnetotellurics survey result, and use 3D first-arrival traveltime tomography to refine the initial velocity model. We then derive a set of elastic wave velocities and anisotropic parameters using elastic full-waveform inversion, and obtain PP and PS images using elastic reverse-time migration. We identify major faults by analyzing the variations of seismic velocities and anisotropy parameters, and reveal mid- to small-scale faults by applying a supervised machine learning method to the seismic migration images. Our characterization reveals complex velocity heterogeneities and anisotropies, as well as faults, with a high spatial resolution. These results can provide valuable information for optimal placement of future injection and production wells to increase geothermal energy production at the Blue Mountain geothermal power plant.

Effects of #coronavirus content moderation on misinformation and anti-Asian hate on Instagram

This study evaluated the intended and unintended effects of Instagram’s content moderation on #coronavirus for both the short- and long-term effects on misinformation and anti-Asian sentiment. We performed manual coding of images ( N = 9648), and a series of supervised machine learning methods to classify three waves of comments ( N = 22,676) published in 2020 on Instagram. Welch’s F tests were used to compare misinformation, emotions, toxicity, and identity attack across three time periods. The results showed that hashtag moderation had an intended effect in reducing misinformation, and an unintended effect in reducing anger, fear, toxicity, and identity attack. Images with people of East Asian descent were associated with more anger, fear, toxicity, and identity attack than images with people of other races. Prior to content moderation, misinformation was associated with identity attack. Stigmatization on social media, and content moderation of misinformation and hate speech are discussed.

Direct and indirect simulating and projecting hydrological drought using a supervised machine learning method

There is a trend in using Artificial Intelligence methods as simulation tools in different aspects of hydrology, including river discharge simulations, drought predictions, and crop yield simulations. The motivation of this work was to assess two various concepts in applying these methods in simulations and projections of hydrological drought. In this study, Standardized Runoff Index (SRI) was simulated and projected using Artificial Neural Networks (ANNs). Maximum and minimum temperature, precipitation, and meteorological drought indicators (the Standardized Precipitation Index (SPI)) were selected as predictors. A direct approach (directly simulating and projecting SRI) and an indirect approach (simulating and projecting river discharge, then calculating SRI) were assessed. Our results show that the indirect approach performs better than the direct approach in simulations of SRI in four discharge stations in the Odra River Basin (a transboundary river basin in Central Europe) from 2000 to 2019. Moreover, a considerable difference between these two approaches was detected in projections of hydrological drought under the RCP8.5 emission scenario for two horizons (near future: 2021–2040, and far future: 2041–2060). Based on the run theory, both approaches show somewhat similar drought conditions for future projections.

Online detection of compensatory strategies in human movement with supervised classification: a pilot study.

Stroke survivors often compensate for the loss of motor function in their distal joints by altered use of more proximal joints and body segments. Since this can be detrimental to the rehabilitation process in the long-term, it is imperative that such movements are indicated to the patients and their caregiver. This is a difficult task since compensation strategies are varied and multi-faceted. Recent works that have focused on supervised machine learning methods for compensation detection often require a large training dataset of motions with compensation location annotations for each time-step of the recorded motion. In contrast, this study proposed a novel approach that learned a linear classifier from energy-based features to discriminate between healthy and compensatory movements and identify the compensating joints without the need for dense and explicit annotations. Six healthy physiotherapists performed five different tasks using healthy movements and acted compensations. The resulting motion capture data was transformed into joint kinematic and dynamic trajectories. Inspired by works in bio-mechanics, energy-based features were extracted from this dataset. Support vector machine (SVM) and logistic regression (LR) algorithms were then applied for detection of compensatory movements. For compensating joint identification, an additional condition enforcing the independence of the feature calculation for each observable degree of freedom was imposed. Using leave-one-out cross validation, low values of mean brier score (<0.15), mis-classification rate (<0.2) and false discovery rate (<0.2) were obtained for both SVM and LR classifiers. These methods were found to outperform deep learning classifiers that did not use energy-based features. Additionally, online classification performance by our methods were also shown to outperform deep learning baselines. Furthermore, qualitative results obtained from the compensation joint identification experiment indicated that the method could successfully identify compensating joints. Results from this study indicated that including prior bio-mechanical information in the form of energy based features can improve classification performance even when linear classifiers are used, both for offline and online classification. Furthermore, evaluation compensation joint identification algorithm indicated that it could potentially provide a straightforward and interpretable way of identifying compensating joints, as well as the degree of compensation being performed.

Cementron: Machine learning the alite and belite phases in cement clinker from optical images

The performance of cement hydrate depends on constituent phases, viz., alite, belite, aluminate, and ferrites present in cement clinker. Traditionally, clinker phases are analyzed from the optical image using empirical image processing techniques. However, the non-uniformity of images, variations in geometry, size of phases, experimental approaches, and imaging methods make it challenging to identify individual phases. Here, we present a machine learning (ML) approach to automatically and accurately detect the major clinker microstructure phases, namely, alite and belite, from optical microscopy images. To this extent, we create an annotated dataset of cement clinker by manually labelling alite and belite phases in microscopy images. We then demonstrate the use of supervised ML methods to train models for identifying alite and belite regions by using a small fraction of annotations on a single image. Further, we finetune the image detection and segmentation Mask R-CNN model using Detectron-2 by considering all the annotated images of the cement microstructure to develop a model for detecting cement phases, namely, Cementron. We demonstrate that Cementron, trained on literature data, generalized well on completely new data obtained from in-house experiments, demonstrating its wider applicability.

Application of Machine Learning Ensemble Methods to ASTRI Mini-Array Cherenkov Event Reconstruction

The Imaging Atmospheric Cherenkov technique has opened up previously unexplored windows for the study of astrophysical radiation sources in the very high-energy (VHE) regime and is playing an important role in the discovery and characterization of VHE gamma-ray emitters. However, even for the most powerful sources, the data collected by Imaging Atmospheric Cherenkov Telescopes (IACTs) are heavily dominated by the overwhelming background due to cosmic-ray nuclei and cosmic-ray electrons. As a result, the analysis of IACT data necessitates the use of a highly efficient background rejection technique capable of distinguishing a gamma-ray induced signal through identification of shape features in its image. We present a detailed case study of gamma/hadron separation and energy reconstruction. Using a set of simulated data based on the ASTRI Mini-Array Cherenkov telescopes, we have assessed and compared a number of supervised Machine Learning methods, including the Random Forest method, Extra Trees method, and Extreme Gradient Boosting (XGB). To determine the optimal weighting for each method in the ensemble, we conducted extensive experiments involving multiple trials and cross-validation tests. As a result of this thorough investigation, we found that the most sensitive Machine Learning technique applied to our data sample for gamma/hadron segregation is a Stacking Ensemble Method composed of 42% Extra Trees, 28% Random Forest, and 30% XGB. In addition, the best-performing technique for energy estimation is a different Stacking Ensemble Method composed of 45% XGB, 27.5% Extra Trees, and 27.5% Random Forest. These optimal weightings were derived from extensive testing and fine-tuning, ensuring maximum performance for both gamma/hadron separation and energy estimation.

A New Remote Hyperspectral Imaging System Embedded on an Unmanned Aquatic Drone for the Detection and Identification of Floating Plastic Litter Using Machine Learning

This paper presents a new Remote Hyperspectral Imaging System (RHIS) embedded on an Unmanned Aquatic Drone (UAD) for plastic detection and identification in coastal and freshwater environments. This original system, namely the Remotely Operated Vehicle of the University of Littoral Côte d’Opale (ROV-ULCO), works in a near-field of view, where the distance between the hyperspectral camera and the water surface is about 45 cm. In this paper, the new ROV-ULCO system with all its components is firstly presented. Then, a hyperspectral image database of plastic litter acquired with this system is described. This database contains hyperspectral data cubes of different plastic types and polymers corresponding to the most-common plastic litter items found in aquatic environments. An in situ spectral analysis was conducted from this benchmark database to characterize the hyperspectral reflectance of these items in order to identify the absorption feature wavelengths for each type of plastic. Finally, the ability of our original system RHIS to automatically recognize different types of plastic litter was assessed by applying different supervised machine learning methods on a set of representative image patches of marine litter. The obtained results highlighted the plastic litter classification capability with an overall accuracy close to 90%. This paper showed that the newly presented RHIS coupled with the UAD is a promising approach to identify plastic waste in aquatic environments.

Robust Malware identification via deep temporal convolutional network with symmetric cross entropy learning

AbstractRecent developments in the field of Internet of things (IoT) have aroused growing attention to the security of smart devices. Specifically, there is an increasing number of malicious software (Malware) on IoT systems. Nowadays, researchers have made many efforts concerning supervised machine learning methods to identify malicious attacks. High‐quality labels are of great importance for supervised machine learning, but noises widely exist due to the non‐deterministic production environment. Therefore, learning from noisy labels is significant for machine learning‐enabled Malware identification. In this study, motivated by the symmetric cross entropy with satisfactory noise robustness, the authors propose a robust Malware identification method using temporal convolutional network (TCN). Moreover, word embedding techniques are generally utilised to understand the contextual relationship between the input operation code (opcode) and application programming interface function names. Here, considering the numerous unlabelled samples in real‐world intelligent environments, the authors pre‐train the TCN model on an unlabelled set using a word embedding method, that is, Word2Vec. In the experiments, the proposed method is compared with several traditional statistical methods and more recent neural networks on a synthetic Malware dataset and a real‐world dataset. The performance comparisons demonstrate the better performance and noise robustness of their proposed method, especially that the proposed method can yield the best identification accuracy of 98.75% in real‐world scenarios.

Intelligent Decision-Making Framework for Evaluating and Benchmarking Hybridized Multi-Deep Transfer Learning Models: Managing COVID-19 and Beyond

In this study, we developed a novel multi-criteria decision-making (MCDM) framework for evaluating and benchmarking hybrid multi-deep transfer learning models using radiography X-ray coronavirus disease (COVID-19) images. First, we collected and pre-processed eight public databases related to the targeted datasets. Second, convolutional neural network (CNN) models extracted features from 1,338 chest X-ray (CXR) frontal view image data using six pre-trained models: VGG16, VGG19, painters, SqueezeNet, DeepLoc, and Inception v3. Then, we used the intersection between the six CNN models and eight classical machine learning (ML) methods, including AdaBoost, Decision Tree, logistic regression, random forest, kNN, neural network, and Naive Bayes, to introduce 48 hybrid classification models. In this study, eight supervised ML methods were used to classify COVID-19 CXR images. The classifiers were implemented using the TensorFlow2 and Keras libraries in Python. A feature vector was extracted from each image, and a five-fold cross-validation technique was used to evaluate the performance. The cost parameter [Formula: see text] was set to 1 and the gamma parameter [Formula: see text] was set to 0.1 for all classifiers. The experiments were run on a Windows-based computer with dual Intel I CoITM i7 processors at 2.50[Formula: see text]GHz, 8[Formula: see text]GB of RAM, and a graphical processing unit of 2[Formula: see text]GB. The performance metrics of the 48 hybrid models, including the classification accuracy (CA), specificity, area under the curve (AUC), F1 score, precision, recall, and log loss, were used as efficient evaluation criteria. Third, the MCDM approach was used, which included (i) developing a dynamic decision matrix based on seven evaluation metrics and the developed hybrid models, (ii) developing the fuzzy-weighted zero-inconsistency method for determining the weight coefficients for the seven-evaluation metrics with zero inconsistency, and (iii) developing the Višekriterijumsko Kompromisno Rangiranje method for benchmarking the 48 hybrid models. Our experimental results reveal that (i) CA and AUC obtained the highest importance weights of 0.164 and 0.147, respectively, whereas F1 and specificity obtained the lowest weights of 0.134 and 0.134, respectively, and (ii) the highest three hybrid model scores were painters neural network, painters logistic regression, and VGG16-logistic regression, making them the highest ranking scores. Finally, the developed framework was validated using sensitivity analysis and comparison analysis.

Health Monitoring System Using Machine Learning Techniques Algorithm

Abstract: One of the most crucial aspects of someone's capacity to progress in life is their physical and mental well-being. Given the resources and requirements of society, the health-care system seeks to improve the populace as effectively as feasible. Due to a lack of timely medical equipment and treatments, death rates are growing in most nations. These health concerns can be prevented by offering standard healthcare services. The Flask framework was used to create the web application that houses our health monitoring system. In this Health Monitoring System, we employed Decision Tree Classification (Supervised Machine Learning method) to precisely anticipate outcomes. We used our own dataset to train and test our model. We could anticipate the patient's health level and area of risk based on that evaluation

Investigating major sources of air pollution and improving spatiotemporal forecast accuracy using supervised machine learning and a proxy

Because weather can’t be controlled, air pollution can only be managed by planning and controlling the sources. Therefore, mathematical models that do not relate the source to the air quality do not offer guidance on how to regulate the air quality. This is often due to a lack of information about sources or a limited budget to measure emissions from sources that are expected to have a major contribution. A proxy for emission rate was used in modeling to improve the prediction of pollutant concentration and to investigate major sources. The daily average air concentration of NO2 and SO2 were predicted at five different locations using supervised machine learning methods, including random forests and support vector machines regression. Three models were constructed using different combinations of features. In model I, only features relating to the weather were used. Model II used only features of the types and quantities of fossil fuels consumed by power plants, which serve as a proxy for emission rates. In model III, all features from models I and II were used. Model III was found to be the most accurate. The average improvement in R2 for NO2 and SO2 regressions for all sites and using both methods were between 61% and 33%, respectively. This improvement in the prediction indicates that power plants are major sources and contribute significantly to the ambient level of SO2 and NO2 levels. The relative importance of all fuel consumption features was found to be comparable to the value of all weather features. The relative importance values of the fuel consumption features show an opportunity to improve air quality by burning a quantity of fuel in a power plant that has a smaller relative importance value instead. However, whether the receptors are located in rural or urban areas must be considered to maximize the net benefits.

530-P: Capillary Blood Ketone Level and the Prediction of Future Diabetic Ketoacidosis Risk in Type 1 Diabetes

Identification of those at risk of diabetic ketoacidosis (DKA) in T1D is challenging. Rather than during illness while DKA is developing, we aimed to determine if levels of routine point-of-care capillary blood ketones could predict DKA months prior to the event. In exploratory analysis, we examined 484 participants randomly assigned to placebo in an adjunct-to-insulin medication trial program (NCT02414958 and NCT02580591). Participants provided morning fasted capillary blood ketone levels twice per week via electronic logbook, and we calculated the maximum and mean level for each individual during a 2-month baseline period. Outcome was the 6 to 12-month occurrence of trial-adjudicated DKA. Area under the Curve (AUC) for the Receiver Operator Characteristic curves were generated. In sensitivity analysis we applied supervised machine learning methods (gradient-boosted trees). Participants had median age 43 [IQR 33, 54], mean HbA1c 8.2±0.6 percent, and provided 2 [1, 4] ketone measurements per week. Twelve DKA events occurred at median 105 [43, 199] days. Maximum and mean ketone levels were higher at baseline for the 12 cases compared to controls (for example, maximum ketone level 0.8 [0.6, 1.2] compared to 0.3 [0.2, 0.7] mmol/L, p=0.002). Maximum ketone level had AUC of 0.769 (95% CI 0.655-0.883). Maximum ketone ≥0.8 mmol/L had sensitivity 64%, specificity 78%, and likelihood ratios positive and negative of 2.9 and 0.5. Machine learning methods outperformed the single-metric analyses. Results provide proof-of-concept that routine capillary ketone surveillance can identify individuals at high-risk of future DKA. Simple or complex machine learning algorithms could be implemented in ketone meters or in future continuous ketone measurement technology. Disclosure C.Song: None. D.Cherney: Other Relationship; Boehringer Ingelheim-Lilly, Merck, AstraZeneca, Sanofi, Mitsubishi-Tanabe, Abbvie, Janssen, Bayer, Prometic, BMS, Maze, Gilead, CSL-Behring, Otsuka, Novartis, Youngene, Lexicon and Novo-Nordisk, Research Support; Boehringer Ingelheim-Lilly, Merck, Janssen, Sanofi, AstraZeneca, CSL-Behring and Novo-Nordisk. D.Mumford: None. G.Tomlinson: None. L.Lovblom: None. B.A.Perkins: Advisory Panel; Dexcom, Inc., Insulet Corporation, Novo Nordisk, Sanofi, Vertex Pharmaceuticals Incorporated, Other Relationship; Abbott, Medtronic, Sanofi, Research Support; Novo Nordisk, Bank of Montreal (BMO). S.Dhaliwal: None. P.Bapat: None. A.M.K.Bakhsh: None. D.R.Budhram: None. D.Scarr: None. A.Weisman: None. M.Fralick: None. N.Ivers: Consultant; Novo Nordisk Canada Inc., IQVIA Inc. Funding Diabetes Canada (OG-3-21-5572-BP)

Fusion of experimental and synthetic data for reliable prediction of steel connection behaviour using machine learning

The development of robust prediction tools based on machine learning (ML) techniques requires the availability of complete, consistent, accurate, and numerous datasets. The application of ML in structural engineering has been limited since, although real size experiments provide complete and accurate data, they are time-consuming and expensive. On the other hand, validated finite element (FE) models provide consistent and numerous synthetic data. Depending on the complexity of the problem, they might require large computational time and cost, and could be subjected to uncertainties and limitation in prediction capability given they are approximations of real-world problems. Hybrid approaches to combine experimental and synthetic datasets have emerged as an alternative to improve the reliability of ML model predictions. In this paper, we explore two hybrid methods to propose a robust approach for the prediction of the extended hollo-bolt (EHB) connection strength, stiffness, and column face displacement: (1) supervised ML methods with data fusion (DF) where learning is optimized with particle swarm optimization (PSO), and (2) artificial neural networks (ANN) based method with model fusion (MF). Based on the analysis of a dataset that combines 22 tensile experimental results with 2000 synthetic datapoints based on FE models, we concluded that using the first method (ML with DF and PSO) is the most suitable method for the prediction of the connection behavior. The ANN-based method with MF shows to be a promising method for the characterization of the EHB connection, however, more extensive experimental data is required for its implementation. Finally, a graphical user interface application was developed and shared in a public repository for the implementation of the proposed hybrid model.

SPAE: Lifelong disk failure prediction via end-to-end GAN-based anomaly detection with ensemble update

Disk failure prediction aims to predict upcoming disk failures in advance for high data reliability. There are numerous supervised machine learning methods that are successful in predicting disk failure using SMART properties as input. However, these approaches heavily rely on a substantial number of annotated failed disks, resulting in degraded prediction performance caused by scarce failed disks at the beginning, also known as the cold start problem. Inspired by the success achieved in Generative Adversarial Network (GAN) based anomaly detection, this paper translates disk failure prediction into an anomaly detection problem. Specifically, we developed a Semi-supervised method for lifelong disk failure Prediction via Adversarial training and Ensemble update, called SPAE. The advantage of SPAE over existing supervised approaches is that SPAE can train the prediction model using only healthy disks, avoiding the cold start problem. Furthermore, SPAE can be updated using ensemble learning on emerging failed disks to resist the model aging problem. Compared to state-of-the-art methods using supervised machine learning on real-world datasets, SPAE predicts disk failures with higher accuracy for the full lifetime of models, i.e., both the startup period and the long-term usage.

Ensemble-based classification approach for PM2.5 concentration forecasting using meteorological data.

Air pollution is a serious challenge to humankind as it poses many health threats. It can be measured using the air quality index (AQI). Air pollution is the result of contamination of both outdoor and indoor environments. The AQI is being monitored by various institutions globally. The measured air quality data are kept mostly for public use. Using the previously calculated AQI values, the future values of AQI can be predicted, or the class/category value of the numeric value can be obtained. This forecast can be performed with more accuracy using supervised machine learning methods. In this study, multiple machine-learning approaches were used to classify PM2.5 values. The values for the pollutant PM2.5 were classified into different groups using machine learning algorithms such as logistic regression, support vector machines, random forest, extreme gradient boosting, and their grid search equivalents, along with the deep learning method multilayer perceptron. After performing multiclass classification using these algorithms, the parameters accuracy and per-class accuracy were used to compare the methods. As the dataset used was imbalanced, a SMOTE-based approach for balancing the dataset was used. Compared to all other classifiers that use the original dataset, the accuracy of the random forest multiclass classifier with SMOTE-based dataset balancing was found to provide better accuracy.

Patterns of valley incision beneath the Greenland Ice Sheet revealed using automated mapping and classification

The Greenland Ice Sheet covers an area of 1.7 million km2, equivalent to ∼79 % of the surface of Kalaallit Nunaat (Greenland) and ∼1.2 % of the Earth's land surface. The macro-scale geomorphology beneath the ice can provide a valuable record of past ice sheet behaviour, particularly during warm periods that may serve as analogues for present and future climates. However, despite extensive mapping of the landscape by airborne radar surveying, Greenland's subglacial geomorphology remains comparatively understudied. Here we construct an automated workflow to identify, extract, and quantify the morphology of valley cross-sectional profiles across Greenland, as observed in NASA Operation IceBridge radar data. We identify 5335 cross-sectional profiles and apply a supervised machine learning method to classify valleys based on their morphological similarity to those formed by glacial or fluvial incision elsewhere in the Northern Hemisphere. Approximately two thirds of the valleys are classified as ‘glacial’, some of which reflect active incision at the modern ice sheet margin, whereas others are situated beneath cold-based, slow-moving ice, indicating that they were incised under a different ice configuration earlier in Greenland's glacial history. The presence of ‘fluvial’ valleys in the low-lying interior of northern Greenland and in mountainous southern Greenland suggests parts of the inherited landscape formed under ice-free conditions during pre- or inter-glacial times have been preserved due to negligible long-term subglacial erosion rates. Some low-lying catchments show hallmarks of a combination of fluvial, glacial, and glacio-fluvial incision, hinting at complex interplays between valley-forming processes over the history of the Greenland Ice Sheet.

TransPhys: Transformer-based unsupervised contrastive learning for remote heart rate measurement

Remote Photoplethysmography (rPPG) is a non-contact heart rate measurement method based on facial video. Because the rPPG signal is highly susceptible to interference from external conditions such as light changes and head movements, the rPPG signal extracted in natural scenes has a low signal to noise ratio and cannot accurately calculate the heart rate value. Aiming at these problems, this paper constructs a self-supervised machine learning network based on Transformer, TransPhys, to achieve robust measurement of heart rate. Using the self-supervised method of contrastive learning, first perform data augmentation, input the enhanced frame into Stem, and extract coarse local spatial features. Through Spatial-Temporal Transformer, calculate the pulse signal of the face area, and finally output the PPG signal. Compared with other supervised machine learning methods, this method can be trained and achieve ideal results without using any labels, which means that this method can be well generalized to practical applications. The experimental results show that this method is better than the current mainstream self-supervised methods.