Current Machine Learning Models Research Articles

PrognosiT: Pathway/gene set-based tumour volume prediction using multiple kernel learning

BackgroundIdentification of molecular mechanisms that determine tumour progression in cancer patients is a prerequisite for developing new disease treatment guidelines. Even though the predictive performance of current machine learning models is promising, extracting significant and meaningful knowledge from the data simultaneously during the learning process is a difficult task considering the high-dimensional and highly correlated nature of genomic datasets. Thus, there is a need for models that not only predict tumour volume from gene expression data of patients but also use prior information coming from pathway/gene sets during the learning process, to distinguish molecular mechanisms which play crucial role in tumour progression and therefore, disease prognosis.ResultsIn this study, instead of initially choosing several pathways/gene sets from an available set and training a model on this previously chosen subset of genomic features, we built a novel machine learning algorithm, PrognosiT, that accomplishes both tasks together. We tested our algorithm on thyroid carcinoma patients using gene expression profiles and cancer-specific pathways/gene sets. Predictive performance of our novel multiple kernel learning algorithm (PrognosiT) was comparable or even better than random forest (RF) and support vector regression (SVR). It is also notable that, to predict tumour volume, PrognosiT used gene expression features less than one-tenth of what RF and SVR algorithms used.ConclusionsPrognosiT was able to obtain comparable or even better predictive performance than SVR and RF. Moreover, we demonstrated that during the learning process, our algorithm managed to extract relevant and meaningful pathway/gene sets information related to the studied cancer type, which provides insights about its progression and aggressiveness. We also compared gene expressions of the selected genes by our algorithm in tumour and normal tissues, and we then discussed up- and down-regulated genes selected by our algorithm while learning, which could be beneficial for determining new biomarkers.

A Novel Deep Learning Approach for Detection of Pneumonia from Chest X-Rays

Current machine learning models for pneumonia detection perform poorly compared to those for other diseases. I aim to use a novel approach to build a model that can improve on performance and be applicable to other diseases.
 My approach uses multi-tier neural networks instead of using single convolutional neural networks (CNNs) like what other researchers have done. My multi-tier model consists of three tier-1 Neural Networks (NNs) and one tier-2 NN. For the tier-1 NN, I use 3 selected ImageNet-trained CNNs (ResNet152V2, DenseNet201, and NASNet Hub) as the starting bases, and train each of them via transfer learning. Then, a tier-2 NN is employed to combine the prediction results from the well-trained tier-1 NNs. The tier-2 NN model is trained with the same dataset. It produces the final predictions with substantially improved performance. The dataset used for my model is pre-processed by me, and based on a public chest X-ray dataset from the NIHCC.
 My model achieved an AUC score of 76.5%, which is better than each of the tier-1 NNs alone and better than most existing models created by others.
 My multi-tier approach accurately detects pneumonia from chest X-rays. It’s practical and it employs incremental learning, meaning it can be continuously improved over time. In the future, I could extend my model from binary classification to multiclass classification, apply it for lung cancer, or even other diseases.

Comparison of machine and deep learning for the classification of cervical cancer based on cervicography images

Cervical cancer is the second most common cancer in women worldwide with a mortality rate of 60%. Cervical cancer begins with no overt signs and has a long latent period, making early detection through regular checkups vitally immportant. In this study, we compare the performance of two different models, machine learning and deep learning, for the purpose of identifying signs of cervical cancer using cervicography images. Using the deep learning model ResNet-50 and the machine learning models XGB, SVM, and RF, we classified 4119 Cervicography images as positive or negative for cervical cancer using square images in which the vaginal wall regions were removed. The machine learning models extracted 10 major features from a total of 300 features. All tests were validated by fivefold cross-validation and receiver operating characteristics (ROC) analysis yielded the following AUCs: ResNet-50 0.97(CI 95% 0.949–0.976), XGB 0.82(CI 95% 0.797–0.851), SVM 0.84(CI 95% 0.801–0.854), RF 0.79(CI 95% 0.804–0.856). The ResNet-50 model showed a 0.15 point improvement (p < 0.05) over the average (0.82) of the three machine learning methods. Our data suggest that the ResNet-50 deep learning algorithm could offer greater performance than current machine learning models for the purpose of identifying cervical cancer using cervicography images.

Unified framework for quantum classification

Quantum machine learning is an emerging field that combines machine learning with advances in quantum technologies. Many works have suggested great possibilities of using near-term quantum hardware in supervised learning. Motivated by these developments, we present an embedding-based framework for supervised learning with trainable quantum circuits. We introduce both explicit and implicit approaches. The aim of these approaches is to map data from different classes to separated locations in the Hilbert space via the quantum feature map. We will show that the implicit approach is a generalization of a recently introduced strategy, so-called \textit{quantum metric learning}. In particular, with the implicit approach, the number of separated classes (or their labels) in supervised learning problems can be arbitrarily high with respect to the number of given qubits, which surpasses the capacity of some current quantum machine learning models. Compared to the explicit method, this implicit approach exhibits certain advantages over small training sizes. Furthermore, we establish an intrinsic connection between the explicit approach and other quantum supervised learning models. Combined with the implicit approach, this connection provides a unified framework for quantum supervised learning. The utility of our framework is demonstrated by performing both noise-free and noisy numerical simulations. Moreover, we have conducted classification testing with both implicit and explicit approaches using several IBM Q devices.

ExpressGesture: Expressive gesture generation from speech through database matching

AbstractCo‐speech gestures are a vital ingredient in making virtual agents more human‐like and engaging. Automatically generated gestures based on speech‐input often lack realistic and defined gesture form. We present a database‐driven approach guaranteeing defined gesture form. We built a large corpus of over 23,000 motion‐captured co‐speech gestures and select individual gestures based on expressive gesture characteristics that can be estimated from speech audio. The expressive parameters are gesture velocity and acceleration, gesture size, arm swivel, and finger extension. Individual, parameter‐matched gestures are then combined into animated sequences. We evaluate our gesture generation system in two perceptual studies. The first study compares our method to the ground truth gestures as well as mismatched gestures. The second study compares our method to five current generative machine learning models. Our method outperformed mismatched gesture selection in the first study and showed competitive performance in the second.

“Call me sexist, but...” : Revisiting Sexism Detection Using Psychological Scales and Adversarial Samples

Research has focused on automated methods to effectively detect sexism online. Although overt sexism seems easy to spot, its subtle forms and manifold expressions are not. In this paper, we outline the different dimensions of sexism by grounding them in their implementation in psychological scales. From the scales, we derive a codebook for sexism in social media, which we use to annotate existing and novel datasets, surfacing their limitations in breadth and validity with respect to the construct of sexism. Next, we leverage the annotated datasets to generate adversarial examples, and test the reliability of sexism detection methods. Results indicate that current machine learning models pick up on a very narrow set of linguistic markers of sexism and do not generalize well to out-of-domain examples. Yet, including diverse data and adversarial examples at training time results in models that generalize better and that are more robust to artifacts of data collection. By providing a scale-based codebook and insights regarding the shortcomings of the state-of-the-art, we hope to contribute to the development of better and broader models for sexism detection, including reflections on theory-driven approaches to data collection.

Towards the Interpretability of Machine Learning Predictions for Medical Applications Targeting Personalised Therapies: A Cancer Case Survey.

Artificial Intelligence is providing astonishing results, with medicine being one of its favourite playgrounds. Machine Learning and, in particular, Deep Neural Networks are behind this revolution. Among the most challenging targets of interest in medicine are cancer diagnosis and therapies but, to start this revolution, software tools need to be adapted to cover the new requirements. In this sense, learning tools are becoming a commodity but, to be able to assist doctors on a daily basis, it is essential to fully understand how models can be interpreted. In this survey, we analyse current machine learning models and other in-silico tools as applied to medicine—specifically, to cancer research—and we discuss their interpretability, performance and the input data they are fed with. Artificial neural networks (ANN), logistic regression (LR) and support vector machines (SVM) have been observed to be the preferred models. In addition, convolutional neural networks (CNNs), supported by the rapid development of graphic processing units (GPUs) and high-performance computing (HPC) infrastructures, are gaining importance when image processing is feasible. However, the interpretability of machine learning predictions so that doctors can understand them, trust them and gain useful insights for the clinical practice is still rarely considered, which is a factor that needs to be improved to enhance doctors’ predictive capacity and achieve individualised therapies in the near future.

Learning framework for carbon emissions predictions incorporating a RReliefF driven features selection and an iterative neural network architecture improvement

Inaccurate carbon emissions predictions may be one of the root factors leading to the overall ineffectiveness of the European Union environmental regulatory framework. Therefore, the present article aims at introducing a novel computational learning framework for carbon emissions prediction incorporating a RReliefF driven features selection and an iterative neural network architecture improvement. Our learning framework algorithmic architecture iteratively chains the features selection process and the backpropagation artificial neural network architecture design based on the data assessment accomplished by the RReliefF algorithm. Thus a better features set / neural network architecture combination is obtained for each specific prediction target. The implemented framework was trained and tested with real world data obtained from the European Union, International Energy Agency, Organisation for Economic Co-operation and Development, and World Bank, for the period 1990–2017. The framework evaluation against current mainstream machine learning models, and its benchmarking comparing to recent published researches on carbon emissions prediction indicates that our research contribution is relevant and capable of supporting the improvement of environmental policies.Graphic abstract

A Tibetan Thangka data set and relative tasks

Data set of high quality is the cornerstone of the current data-driven machine learning models, and plays an important role in promoting the development of various application areas. At present, image analysis and processing techniques have intensively involved into the tasks of inheriting and protecting culture resources. However, currently there are few effective image data sets about the traditional Chinese Tibetan culture. In this work, we provided a small data set referred as CYTKv1(Chomo Yarlung Tibet version 1) which includes 1700+ Thangka images (an important and representative carrier of Chinese Tibetan culture), and the main objects in each image are manually labeled and bounding-boxed with semantic words. In addition, we shared a list of tasks of processing and analyzing the Thangkas to enlighten researchers about the challenges and potential applications on this data set. At last, we tested several famous deep learning models for the purpose of validating the annotation task on the new data set and presented the results of them, and finally selected the best one as the baseline for the annotation task.

An Alternative Method for Traffic Accident Severity Prediction: Using Deep Forests Algorithm

Traffic safety has always been an important issue in sustainable transportation development, and the prediction of traffic accident severity remains a crucial challenging issue in the domain of traffic safety. A huge variety of forecasting models have been proposed to meet this challenge. These models gradually evolved from linear to nonlinear forms and from traditional statistical regression models to current popular machine learning models. Recently, a machine learning algorithm called Deep Forests based on the decision tree ensemble has aroused widespread concern, which was proposed for the first time by a research team of Nanjing University. This algorithm was proved to be more accurate and robust in comparison with other machine learning algorithms. Motivated by this benefit, this study employs the UK road safety dataset to propose a novel method for predicting the severity of traffic accidents based on the Deep Forests algorithm. To verify the superiority of our proposed method, several other machine learning algorithm-based perdition models were implemented to predict traffic accident severity with the same dataset, and the prediction results show that the Deep Forests algorithm present good stability, fewer hyper-parameters, and the highest accuracy under different level of training data volume. It is expected that the findings from this study would be helpful for the establishment or improvement of effective traffic safety system within a sustainable transportation system, which is of great significance for helping government managers to establish timely proactive strategies in traffic accident prevention and effectively improve road traffic safety.

Lost in Translation? From Conventional Scoring Tools to Modern Data-Driven Risk Assessment in Critical Care Medicine

High-resolution, longitudinal health data became widely available in intensive care units in the past years. Patient risk assessment, however, is still primarily based on conventional scores that take into account only a few parameters taken at single time points, which frequently causes inaccurate predictions in the clinical practice. Likewise, the contribution of AI-approaches remains sparse, as current machine learning models are inherently difficult to deduce and even impressive results rarely contribute to disease understanding. This review focusses on the limitations of conventional risk scores, and on recent developments and challenges of novel, data-driven assessment tools.

ESTIMATING CROP YIELDS WITH REMOTE SENSING AND DEEP LEARNING

Abstract. Increasing the accuracy of crop yield estimates may allow improvements in the whole crop production chain, allowing farmers to better plan for harvest, and for insurers to better understand risks of production, to name a few advantages. To perform their predictions, most current machine learning models use NDVI data, which can be hard to use, due to the presence of clouds and their shadows in acquired images, and due to the absence of reliable crop masks for large areas, especially in developing countries. In this paper, we present a deep learning model able to perform pre-season and in-season predictions for five different crops. Our model uses crop calendars, easy-to-obtain remote sensing data and weather forecast information to provide accurate yield estimates.

Reducing IMRT QA Workload by 95% and Keeping the Same Level of Quality Control

Reducing the maximum possible error (max-error) in a quality assurance program is the natural metric in contrast to reducing the mean value of the expected error. Current machine learning (ML) models used to predict IMRT QA, however, minimize the later and not the former. Here, we propose a novel ML algorithm for predicting QA plan passing rate that minimizes the max-error. From a single institution, 498 patient IMRT QA plans were measured using a commercial 2D diode array (Sun Nuclear’s MapCheck®) and delivered on five isocentric linear accelerators. Passing rates were determined based on 3%/3mm local dose and distance-to-agreement. Each treatment plan was characterized by 78 features to quantify complexity, such as MLC position, field area, X/Y jaw position, etc. Two ML models were used to predict passing rate: (1) an ordinary least squares (OLS) linear model that minimize the mean of the expected error; and (2) Chebyshev minimax (MM) linear solution that minimizes max-error. An 80% training dataset was used for fitting and optimization, and a 20% testing dataset was used to compare predictions. Model fitting and optimization (using Ridge, Lasso and Lasso-like regularization) was performed in a computer algorithm. When using all 498 plans for training, the max-error of IMRT QA passing rate predictions was 7.6% for the OLS model and 3.0% for the MM model. The mean square error was, however, 1.4% and 2.8% for the OLS and MM models respectively showing the tradeoff between both optimization methods. Following optimization and testing with a dataset that the model had not seen before (20% plans), the OLS and MM model max-errors were 7.0% and 3.8% respectively. To reach the clinical goal that all plans have at least a 90% passing rate, using the OLS we would have to QA all plans predicted to be 97.0% passing rate or lower (79 out of 497, 16.0%). Whereas with the MM model, we would only need to QA all plans predicted to be 93.8% or lower (24 out of 497, 4.8%) which is a 4x reduction compared to the OLS model. Efficient and safe QA programs for IMRT treatments require accurate identification of plans that are likely to not meet passing criteria. Such a program would enable relocation of clinical resources to higher priority tasks, which is in alignment with TG100 guidelines. Using Chebyshev minimax optimization, we could reduce resources for IMRT QA by 95% while guaranteeing a safe clinical QA program that meets current passing criteria.

Prediction models for COVID-19 clinical decision making

Prediction models for COVID-19 clinical decision making

A hybrid diagnosis method for inverter open-circuit faults in PMSM drives

In order to improve the evaluation process of inverter open-circuit faults diagnosis in permanent magnet synchronous motor (PMSM) drives, this paper presents a diagnosis method based on current residuals and machine learning models. The machine learning models are introduced to make a comprehensive evaluation for the current residuals obtained from a state observer, instead of evaluating the residuals by comparing with thresholds. Meanwhile, fault diagnosis and location are conducted simultaneously by the machine learning models, which simplifies the diagnosis process. Besides, a sampling strategy is designed to implement the proposed scheme online. Experiments are carried out on a DSP based PMSM drive, and the effectiveness of the proposed method is verified.

Artificial Intelligence in plastic surgery: What is it? Where are we now? What is on the horizon?

An increasing quantity of data is required to guide precision medicine and advance future healthcare practices, but current analytical methods often become overwhelmed. Artificial intelligence (AI) provides a promising solution. Plastic surgery is an innovative surgical specialty expected to implement AI into current and future practices. It is important for all plastic surgeons to understand how AI may affect current and future practice, and to recognise its potential limitations. Peer-reviewed published literature and online content were comprehensively reviewed. We report current applications of AI in plastic surgery and possible future applications based on published literature and continuing scientific studies, and detail its potential limitations and ethical considerations. Current machine learning models using convolutional neural networks can evaluate breast mammography and differentiate benign and malignant tumours as accurately as specialist doctors, and motion sensor surgical instruments can collate real-time data to advise intraoperative technical adjustments. Centralised big data portals are expected to collate large datasets to accelerate understanding of disease pathogeneses and best practices. Information obtained using computer vision could guide intraoperative surgical decisions in unprecedented detail and semi-autonomous surgical systems guided by AI algorithms may enable improved surgical outcomes in low- and middle-income countries. Surgeons must collaborate with computer scientists to ensure that AI algorithms inform clinically relevant health objectives and are interpretable. Ethical concerns such as systematic biases causing non-representative conclusions for under-represented patient groups, patient confidentiality and the limitations of AI based on the quality of data input suggests that AI will accompany the plastic surgeon, rather than replace them.

SO019A PREDICTIVE MODEL OF MORTALITY IN ACUTE RENAL FAILURE IN THE CRITICAL PATIENT: USEFULNESS OF ARTIFICIAL INTELLIGENCE

Abstract Background and Aims Patients with Acute Renal Failure (ARF) have a high risk of mortality, especially those who enter the Intensive Care Unit (ICU). In this population, predictive models of mortality on prognostic scales, such as SAPS-II (Simplified Acute Physiology Score II), linearly relate risk factors without taking into account the complex relationship's variables can have. There are models where Machine Learning (ML) techniques have been used, but there is still room for improvement. The implementation of deep artificial neural networks (DANN) can be challenging. The literature models, using SAPS-II report an accuracy, f1 and ROC area (receiver operating curve) in ranges of 0.538-0.621, 0.333-0.377 and 0.781-0.809 respectively. The best results with ML are improved in neural networks of a hidden layer or random forest, being the best performance in the latter: accuracy 0.715-0.741, F1 0.449-0.470 and ROC between 0.862-0.870. The aim is to evaluate and improve the predictive capacity of ML techniques for the prediction of mortality in patients with ARF admitted to the ICU, through the use of the open database MIMIC-III (Medical Information Martfor Intensive Care III). Method Design: Retrospective analysis of historical cooperation of 20,928 patients with ARF from Beth Israel Deaconess (Boston), from 2001 to 2012. Method: ML algorithm based on DANN. Creation of a model to predict in-hospital mortality after discharge from the ICU with the variables of the first 24 hours after admission to the ICU. To evaluate the robustness of the model has been performed cross-validation by separating the samples into different combinations of training and test data (k-folds). The unavailable variables haematological with means extracted from the training set of the respective fold. The DANN trained with the complex relationship's variables folds, two hidden layers of 75 and 40 neurons respectively. Inclusion criteria: &amp;gt; 16 years, AKI according to KDIGO criteria and predictive variables based on SAPS-II. Variables: Age, sex, type of admission, number of admissions in ICU, heart rate, blood pressure, temperature, PaO2, FiO2, sodium, potassium, bilirubin, bicarbonate, urea, leukocytes, diuresis and diagnosis of ICD-9 (metastatic cancer, haematological malignancies). Results Accuracy: 76.7% 0 ± 1.7%, f1: 86.6% ± 1.9%, RO area: 0.859 ± 0.006, sensitivity: 75.7% ± 2.8%, specificity: 80.5% ± 2.9%. Conclusion The use of Machine Learning techniques based on deep artificial neural networks can improve the predictive ability of mortality in acute renal failure of the critically ill patient of traditional clinical risk scales and even current Machine Learning models.

Kernel-Based Machine Learning for Efficient Simulations of Molecular Liquids.

Current machine learning (ML) models aimed at learning force fields are plagued by their high computational cost at every integration time step. We describe a number of practical and computationally efficient strategies to parametrize traditional force fields for molecular liquids from ML: the particle decomposition ansatz to two- and three-body force fields, the use of kernel-based ML models that incorporate physical symmetries, the incorporation of switching functions close to the cutoff, and the use of covariant meshing to boost the training set size. Results are presented for model molecular liquids: pairwise Lennard-Jones, three-body Stillinger–Weber, and bottom-up coarse-graining of water. Here, covariant meshing proves to be an efficient strategy to learn canonically averaged instantaneous forces. We show that molecular dynamics simulations with tabulated two- and three-body ML potentials are computationally efficient and recover two- and three-body distribution functions. Many-body representations, decomposition, and kernel regression schemes are all implemented in the open-source software package VOTCA.

Toward automatic comparison of visualization techniques: Application to graph visualization

Many end-user evaluations of data visualization techniques have been run during the last decades. Their results are cornerstones to build efficient visualization systems. However, designing such an evaluation is always complex and time-consuming and may end in a lack of statistical evidence and reproducibility. We believe that modern and efficient computer vision techniques, such as deep convolutional neural networks (CNNs), may help visualization researchers to build and/or adjust their evaluation hypothesis. The basis of our idea is to train machine learning models on several visualization techniques to solve a specific task. Our assumption is that it is possible to compare the efficiency of visualization techniques based on the performance of their corresponding model. As current machine learning models are not able to strictly reflect human capabilities, including their imperfections, such results should be interpreted with caution. However, we think that using machine learning-based pre-evaluation, as a pre-process of standard user evaluations, should help researchers to perform a more exhaustive study of their design space. Thus, it should improve their final user evaluation by providing it better test cases. In this paper, we present the results of two experiments we have conducted to assess how correlated the performance of users and computer vision techniques can be. That study compares two mainstream graph visualization techniques: node-link (NL) and adjacency-matrix (AM) diagrams. Using two well-known deep convolutional neural networks, we partially reproduced user evaluations from Ghoniem et al. and from Okoe et al.. These experiments showed that some user evaluation results can be reproduced automatically.

Field data–driven online prediction model for icing load on power transmission lines

Methods for the accurate prediction of icing loads in overhead transmission lines have become an important research topic for electrical power systems as they are necessary for ensuring the safety and stability of power-grid operations. Current machine learning models for the prediction of icing loads on transmission lines are afflicted by the following issues: insufficient prediction accuracy, high randomity in the selection of kernel functions and model parameters, and a lack of generalizability. To address these issues, we propose a field data–driven online prediction model for icing loads on transmission lines. First, the effects of the type of kernel function used in the support vector regression algorithm on the prediction accuracy of the model were analyzed using micrometeorological data and icing data collected by on-site monitoring systems. The particle swarm optimization algorithm was then used to optimize and determine the model parameters such as penalty coefficients. An offline support vector regression prediction model was thus constructed. Using the accurate online support vector regression algorithm, the weighting coefficients of the samples were dynamically adjusted to satisfy the Karush–Kuhn–Tucker conditions, which allowed online updates to be made to the regression function and prediction model. Finally, a simulation analysis was performed using actual icing incidents that occurred in a transmission line of the Yunnan Power Grid, which demonstrated that our model can make online predictions for the icing load on transmission lines in actual applications. Our model proved to be superior to conventional icing-load prediction models with regard to the single-step and multi-step prediction accuracies and generalizability. Hence, our prediction model will improve the decision-making processes regarding the deicing and maintenance of power transmission and transformation systems.