Reliable Training Dataset Research Articles

Four-Wheeled Vehicle Sideslip Angle Estimation: A Machine Learning-Based Technique for Real-Time Virtual Sensor Development

In the last few decades, the role of vehicle dynamics control systems has become crucial. In this complex scenario, the correct real-time estimation of the vehicle’s sideslip angle is decisive. Indeed, this quantity is deeply linked to several aspects, such as traction and stability optimization, and its correct understanding leads to the possibility of reaching greater road safety, increased efficiency, and a better driving experience for both autonomous and human-controlled vehicles. This paper aims to estimate accurately the sideslip angle of the vehicle using different neural network configurations. Then, the proposed approach involves using two separate neural networks in a dual-network architecture. The first network is dedicated to estimating the longitudinal velocity, while the second network predicts the sideslip angle and takes the longitudinal velocity estimate from the first network as input. This enables the creation of a virtual sensor to replace the real one. To obtain a reliable training dataset, several test sessions were conducted on different tracks with various layouts and characteristics, using the same reference instrumented vehicle. Starting from the acquired channels, such as lateral and longitudinal acceleration, steering angle, yaw rate, and angular wheel speeds, it has been possible to estimate the sideslip angle through different neural network architectures and training strategies. The goodness of the approach was assessed by comparing the estimations with the measurements obtained from an optical sensor able to provide accurate values of the target variable. The obtained results show a robust alignment with the reference values in a great number of tested conditions. This confirms that the adoption of artificial neural networks represents a reliable strategy to develop real-time virtual sensors for onboard solutions, expanding the information available for controls.

Objective satellite methods including AI algorithms reviewed for the tenth International workshop on tropical cyclones (IWTC-10)

Here we explore the latest four years (2019–2022) of using satellite data to objectively analyze tropical cyclones (TC) and issue recommendations for improved analysis. We first discuss new methods of direct retrieval from SAR and geostationary imagers. Next, we survey some of the most prominent new techniques in AI and discuss their major capabilities (especially accuracy in nonlinear TC behavior, characterization of model uncertainty and creation of synthetic satellite imagery) and limitations (especially lack of transparency and limited amount of training data). We also identify concerns with biases and unlabeled uncertainties in the Best Track records as being a first-order limitation for further progress in objective methods. The article concludes with recommendations to improve future objective methods, especially in the area of more accurate and reliable training data sets.

A Friend in Need is a Friend Indeed: Investigating the Quality of Training Data from Peers for Auto-generating Empathetic Textual Responses to Non-Sensitive Posts in a Cohort of College Students

Towards providing personalized care, digital mental-wellness apps today ask questions to learn about subjects. However, not all subjects using these apps will have mood problems, so they do not need follow-up questions. In this study, we investigate an alternate mechanism to handle such non-sensitive posts (i.e., those not indicating mood problems) in college settings. To do so, we generate and use training data provided by a cohort of peer college students so that responses to non-sensitive posts are contextual, emotionally aware, and empathetic while also being terminal (not asking follow-up questions). Using data from a real mental-wellness app used by students, we identify that AI models trained with our peer-provided dataset generate desirable responses to non-sensitive posts, while models trained with state-of-the-art (Facebook’s) Empathetic Dataset yields responses that ask many follow-up questions, hence giving a perception of being intrusive. We believe that mental wellness apps today must not assume that any subject using these apps has mood problems. Perceptions of intrusiveness (i.e., apps asking many questions) must be a factor in design. We also believe that peer students can provide rich and reliable training datasets for college mental wellness apps, a topic that is not yet explored.

MTCSNet: Mean Teachers Cross-Supervision Network for Semi-Supervised Cloud Detection

Cloud detection methods based on deep learning depend on large and reliable training datasets to achieve high detection accuracy. There will be a significant impact on their performance, however when the training data are insufficient or when the label quality is low. Thus, to alleviate this problem, a semi-supervised cloud detection method, named the mean teacher cross-supervision cloud detection network (MTCSNet) is proposed. This method enforces both consistency and accuracy on two cloud detection student network branches, which are perturbed with different initializations, for the same input image. For each of the two student branches, the respective teacher branches, used to generate high-quality pseudo labels, are constructed using an exponential moving average method (EMA). A pseudo one-hot label, produced by one teacher network branch, supervises the other student network branch with the standard cross-entropy loss, and vice versa. To incorporate additional prior information into the model, the presented method uses near-infrared bands instead of red bands as model inputs and injects strong data augmentations on unlabeled images fed into the student model. This induces the model to learn richer representations and ensure consistency constraints on the predictions of the same unlabeled image across different batches. To attain a more refined equilibrium between the supervised and semi-supervised loss in the training process, the proposed cloud detection network learns the optimal weights based on homoscedastic uncertainty, thus effectively exploiting the advantages of semi-supervised tasks and elevating the overall performance. Experiments on the SPARCS and GF1-WHU public cloud detection datasets show that the proposed method outperforms several state-of-the-art semi-supervised algorithms when only a limited number of labeled samples are available.

Finite strain poro-hyperelasticity: an asymptotic multi-scale ALE-FSI approach supported by ANNs

This contribution introduces and discusses a formulation of poro-hyperelasticity at finite strains. The prediction of the time-dependent response of such media requires consideration of their characteristic multi-scale and multi-physics parameters. In the present work this is achieved by formulating a non-dimensionalised fluid–solid interaction problem (FSI) at the pore level using an arbitrary Lagrange–Euler description (ALE). The resulting coupled systems of PDEs on the reference configuration are expanded and analysed using the asymptotic homogenisation technique. This approach yields three partially novel systems of PDEs: the macroscopic/effective problem and two supplementary microscale problems (fluid and solid). The latter two provide the microscopic response fields whose average value is required in real-time/online form to determine the macroscale response (a concurrent multi-scale approach). In order to overcome the computational challenges related to the above multi-scale closure, this work introduces a surrogate approach for replacing the direct numerical simulation with an artificial neural network. This methodology allows for solving finite strain (multi-scale) porohyperelastic problems accurately using direct automated differentiation through the strain energy. Optimal and reliable training data sets are produced from direct numerical simulations of the fully-resolved problem by including a simple real-time output density check for adaptive sampling step refinement. The data-driven approach is complemented by a sensitivity analysis of the RVE response. The significance of the presented approach for finite strain poro-elasticity/poro-hyperelasticity is shown in the numerical benchmark of a multi-scale confined consolidation problem. Finally, to show the robustness of the method, the system response is dimensionalised using characteristic values of soil and brain mechanics scenarios.

A framework for automatically constructing a dataset for training a vehicle detector

Object detection based on a trained detector has been widely applied to diverse tasks such as pedestrian, face, and vehicle detection. In such approach, detectors are learned offline with an enormous number of training samples. However, the approach has a significant drawback that heavy intervention and effort, as well as domain knowledge, of a human are essentially required to construct a reliable training dataset. To remedy this drawback, we propose a framework to collect and label training samples automatically. By analysing information of foreground blobs obtained from background subtraction results, a training dataset can be constructed without any human's effort. Also, condition investigation of scenes is performed periodically to check the suitability of sample candidates. As a result, it generates an accurate vehicle detector. With the proposed method, training samples can be automatically collected only when vehicle blobs in the given scene provide suitable appearance information. The effectiveness of the proposed framework is demonstrated from vehicle detection tasks under real traffic environments.

A framework for automatically constructing a dataset for training a vehicle detector

Object detection based on a trained detector has been widely applied to diverse tasks such as pedestrian, face, and vehicle detection. In such approach, detectors are learned offline with an enormous number of training samples. However, the approach has a significant drawback that heavy intervention and effort, as well as domain knowledge, of a human are essentially required to construct a reliable training dataset. To remedy this drawback, we propose a framework to collect and label training samples automatically. By analysing information of foreground blobs obtained from background subtraction results, a training dataset can be constructed without any human's effort. Also, condition investigation of scenes is performed periodically to check the suitability of sample candidates. As a result, it generates an accurate vehicle detector. With the proposed method, training samples can be automatically collected only when vehicle blobs in the given scene provide suitable appearance information. The effectiveness of the proposed framework is demonstrated from vehicle detection tasks under real traffic environments.

Quantitative Biomedical Imaging: Techniques and Clinical Applications.

Quantitative Biomedical Imaging: Techniques and Clinical Applications.

Self-Trained Supervised Segmentation of Subcortical Brain Structures Using Multispectral Magnetic Resonance Images.

The aim of this paper is investigate the feasibility of automatically training supervised methods, such as k-nearest neighbor (kNN) and principal component discriminant analysis (PCDA), and to segment the four subcortical brain structures: caudate, thalamus, pallidum, and putamen. The adoption of supervised classification methods so far has been limited by the need to define a representative training dataset, operation that usually requires the intervention of an operator. In this work the selection of the training data was performed on the subject to be segmented in a fully automated manner by registering probabilistic atlases. Evaluation of automatically trained kNN and PCDA classifiers that combine voxel intensities and spatial coordinates was performed on 20 real datasets selected from two publicly available sources of multispectral magnetic resonance studies. The results demonstrate that atlas-guided training is an effective way to automatically define a representative and reliable training dataset, thus giving supervised methods the chance to successfully segment magnetic resonance brain images without the need for user interaction.

Iterative Self-Organizing Atherosclerotic Tissue Labeling in Intravascular Ultrasound Images and Comparison With Virtual Histology

Intravascular ultrasound (IVUS) is the predominant imaging modality in the field of interventional cardiology that provides real-time cross-sectional images of coronary arteries and the extent of atherosclerosis. Due to heterogeneity of lesions and stringent spatial/spectral behavior of tissues, atherosclerotic plaque characterization has always been a challenge and still is an open problem. In this paper, we present a systematic framework from in vitro data collection, histology preparation, IVUS-histology registration along with matching procedure, and finally a robust texture-derived unsupervised atherosclerotic plaque labeling. We have performed our algorithm on in vitro and in vivo images acquired with single-element 40 MHz and 64-elements phased array 20 MHz transducers, respectively. In former case, we have quantified results by local contrasting of constructed tissue colormaps with corresponding histology images employing an independent expert and in the latter case, virtual histology images have been utilized for comparison. We tackle one of the main challenges in the field that is the reliability of tissues behind arc of calcified plaques and validate the results through a novel random walks framework by incorporating underlying physics of ultrasound imaging. We conclude that proposed framework is a formidable approach for retrieving imperative information regarding tissues and building a reliable training dataset for supervised classification and its extension for in vivo applications.

Development of a modified neural network-based land cover classification system using automated data selector and multiresolution remotely sensed data

Integrating multiple images with artificial neural networks (ANN) improves classification accuracy. ANN performance is sensitive to training datasets. Complexity and errors compound when merging multiple data, pointing to needs for new techniques. Kohonen's self-organizing mapping (KSOM) neural network was adapted as an automated data selector (ADS) to replace manual training data processes. The multilayer perceptron (MLP) network was then trained using automatically extracted datasets and used for classification. Two hypotheses were tested: ADS adapted from the KSOM network provides adequate and reliable training datasets, improving MLP classification performance; and fusion of Landsat thematic mapper (TM) and SPOT images using the modified ANN approach increases accuracy. ADS adapted from the KSOM network improved training data quality and increased classification accuracy and efficiency. Fusion of compatible multiple data can improve performance if appropriate training datasets are collected. This proved to be a viable classification scheme particularly where acquiring sufficient and reliable training datasets is difficult.

Improving performance of mammalian microRNA target prediction

BackgroundMicroRNAs (miRNAs) are single-stranded non-coding RNAs known to regulate a wide range of cellular processes by silencing the gene expression at the protein and/or mRNA levels. Computational prediction of miRNA targets is essential for elucidating the detailed functions of miRNA. However, the prediction specificity and sensitivity of the existing algorithms are still poor to generate meaningful, workable hypotheses for subsequent experimental testing. Constructing a richer and more reliable training data set and developing an algorithm that properly exploits this data set would be the key to improve the performance current prediction algorithms.ResultsA comprehensive training data set is constructed for mammalian miRNAs with its positive targets obtained from the most up-to-date miRNA target depository called miRecords and its negative targets derived from 20 microarray data. A new algorithm SVMicrO is developed, which assumes a 2-stage structure including a site support vector machine (SVM) followed by a UTR-SVM. SVMicrO makes prediction based on 21 optimal site features and 18 optimal UTR features, selected by training from a comprehensive collection of 113 site and 30 UTR features. Comprehensive evaluation of SVMicrO performance has been carried out on the training data, proteomics data, and immunoprecipitation (IP) pull-down data. Comparisons with some popular algorithms demonstrate consistent improvements in prediction specificity, sensitivity and precision in all tested cases. All the related materials including source code and genome-wide prediction of human targets are available at http://compgenomics.utsa.edu/svmicro.html.ConclusionsA 2-stage SVM based new miRNA target prediction algorithm called SVMicrO is developed. SVMicrO is shown to be able to achieve robust performance. It holds the promise to achieve continuing improvement whenever better training data that contain additional verified or high confidence positive targets and properly selected negative targets are available.

Establishing a statistic model for recognition of steroid hormone response elements

Identification of hormone response elements (HREs) is essential for understanding the mechanism of hormone-regulated gene expression. To date, there has been a lack of effective bioinformatics tools for recognition of specific HRE such as Progesterone Response Elements (PRE). In this paper, a comprehensive survey and comparison of in silico methods is conducted for establishing a more accurate statistic model. Homogeneity of steroid HRE is analyzed and a reliable training dataset is constructed through extensive searching for experimentally validated response elements from more than 150 literature sources. Based on the observation that the verified HREs carry di-nucleotide preservation in comparison with uniform nucleotide distributions, both mono and di-nucleotide Position Weight Matrices are computed to extract the statistic pattern of the positions. It is followed by the sequence transition pattern recognition using a specifically designed profile Hidden Markov Model. Reciprocal combination of the statistic and transition patterns significantly improves the performance of the model in terms of higher sensitivity and specificity. Upon acquisition of the putative response elements in the promoter areas of vertebrate genes, a qualitative scheme is applied to assess the probability for each gene to be a hormone primary target. Using >650 records of experimentally validated steroid hormone response elements, a high sensitivity level of 73% and high specificity level of one prediction per 8.24 kb is reached, allowing this model to be used for further prediction of primary target genes through the analysis of their upstream promoters, for human or other vertebrate genomes of interest. Additional documents, supplementary data and the web-based program developed for response elements prediction are freely available for academic research at . Submission of putative gene promoter regions for recognition of potential regulatory PREs can be as long as 5 kb.