Generate Training Datasets Research Articles

Synthetic Training Dataset Generation Using a Digital Twin-based Autonomous Driving Simulator

Synthetic Training Dataset Generation Using a Digital Twin-based Autonomous Driving Simulator

Imbalanced ensemble learning leveraging a novel data-level diversity metric

Ensemble learning is one of the best solutions for imbalanced classification problems. Diversity is a key factor that affects the performance of ensemble learning. Most existing diversity metrics such as Q-statistics measure diversity based on the outputs of the base classifiers, incurring high complexity on model training due to the need to re-train base classifiers to achieve satisfactory diversity. We propose a new diversity measure, named Instance Euclidean Distance metric (IED), to evaluate diversity directly based on the training data without training base classifiers, which can significantly cut down the time costs of diversity measuring. A new imbalanced ensemble learning algorithm named P-EUSBagging is proposed to reduce training complexity and improve learning performance by combining IED with population-based incremental learning to generate training datasets with the maximal data-level diversity. Experimental results demonstrate that the diversities measured by IED and three classifier-based diversity measures exhibit a mean absolute correlation coefficient of 0.94, and P-EUSBagging significantly reduces training time and improves learning performance on both Geometric Mean (G-Mean) and Area Under the Curve (AUC).

Adaptive training dataset generation for neural network numerical dispersion mitigation approach in seismic modeling

Adaptive training dataset generation for neural network numerical dispersion mitigation approach in seismic modeling

NIMO: A graphical user interface‐based R package for species distribution modelling

Abstract Species distribution modelling (SDM) is essential for understanding and predicting biodiversity patterns globally. However, the complexities of data preparation, algorithm selection and model evaluation can present significant challenges. This paper introduces NIMO, an R package with a user‐friendly graphical interface aimed at streamlining the entire SDM workflow. In SDM, data preparation is a crucial and time‐consuming process, which requires the acquisition of spatiotemporal occurrence records. The Global Biodiversity Information Facility (GBIF) database plays a key role in addressing data scarcity by providing extensive community science‐based repositories. However, compiling, aligning and verifying spatial references can be laborious. NIMO addresses these challenges by offering a flexible workflow that allows users to more easily query biodiversity data. NIMO organizes its features into three main steps: pre‐modelling, modelling and post‐modelling. It empowers users to optimize data for modelling with a variety of algorithms. NIMO allows for default hyperparameter values or extensive tuning, providing users with flexibility in selecting models. Practical implication: To demonstrate the effectiveness of NIMO, we presented a case study focusing on the critically endangered white‐backed vulture (Gyps africanus) in Kruger National Park, South Africa. We used NIMO to collect GBIF occurrence data and integrated environmental predictors to map suitable habitat (probability of occurrence). The distribution modelling workflow included area calibration, collinearity check, generation of training and testing data sets, model fitting and model combination. The results showcase the discriminative ability of fitted models, especially the generalized linear model, support vector machine and random forest, that constituted the ensemble model. Spatial prediction using the ensemble model revealed that the species has a higher occurrence probability close to roads, indicating a possible increased risk of mortality for the vultures in Kruger National Park from motor vehicle collisions.

Development of a simulation model and adjustment of hyperparameters of a neural network for predicting possible states of a network of small space carriages

The purpose of the research is to develop a simulation model and tune the hyperparameters of a neural network to predict possible states of a network of small spacecraft.The research methods are based on the concepts of the theory of artificial intelligence to control a group of small spacecraft (SSV) – the use of adaptive methods and tools to make decisions, similar to the mechanisms of human thinking. In relation to space communication systems with a heterogeneous structure, artificial intelligence methods and technologies are aimed at predicting the state of communication channels between network nodes and automatically reconfiguring a network of devices based on neural network learning processes. One of the most important functions of network software for the application of cognitive algorithms is to predict the quality of communication between pairs of SSVs.Results. A method has been developed for using the Transformer architecture neural network to predict possible states of the SSV network, which provides aggregation and time synchronization of data on the state of the SSV network, their use for training the neural network, as well as using the neural network to predict the quality of communication. The data format for the training sample has been created, based on the representation of the state of the SSV network, which ensures the generation of the initial state of the network, modeling the proactive mode of its operation, collecting SSV network state markers to generate training data sets in the form of chronological sequences grouped into frames, and allowing to reduce the amount of data transferred between SSVs when creating a training set. A simulation model of the SSV network has been developed, which provides generation of the initial state of the network, modeling of the proactive mode of its operation, and collection of information about the state of the SSV network to generate sets of synthetic training data.Conclusion. The article develops a simulation model of the SSV network for generating synthetic training data and predicting possible states of the SSV network, as well as a method for using a neural network to predict possible states of the SSV network.

Image generation of log ends and patches of log ends with controlled properties using generative adversarial networks

The appearance of the log cross-section provides important information when assessing the quality of the log, where properties to consider include pith location and density of annual rings. This makes tasks like estimation of pith location and annual ring detection of great interest. However, creating labeled training data for these tasks can be time-consuming and subject to misjudgments. For this reason, we aim to create generated training data with controlled properties of pith location and amount of annual rings. We propose a two-step generator based on generative adversarial networks in which we can completely avoid manual labeling, not only when generating training data but also during training of the generator itself. This opens up the possibility to train the generator on other types of log end data without the need to manually label new training data. The same method is used to create two generated training datasets; one of entire log ends and one of patches of log ends. To evaluate how the generated data compares to real data, we train two deep learning models to perform estimation of pith location and ring counting, respectively. The models are trained separately on real and generated data and evaluated on real data only. The results show that the performance of both estimation of pith location and ring counting can be improved by replacing real training data with larger sets of generated training data.

Subspace graph networks for real-time granular flow simulation with applications to machine-terrain interactions

Granular flows interacting with rigid bodies are key to machine-terrain interactions in robotics and related fields, and still pose several open problems. Continuum methods and deep learning, separately and in combination, show promise. This research advances machine learning methods for modeling rigid body-driven granular flows, for terrestrial industrial machine and planetary rover (where gravity is an important factor) applications. The original contribution of this research is the application of a subspace machine learning simulation approach for these engineering problems. To generate training datasets, a high-fidelity continuum method, the material point method (MPM) is utilized. Principal component analysis (PCA) is used to reduce the dimensionality of data. It is shown that the first few (8 in our datasets) principal components of our high-dimensional data keep almost the entire variance in data. A graph network simulator (GNS) is trained to learn the underlying subspace dynamics. The learned GNS is then able to predict particle positions and interaction forces with good accuracy, averaging 6×10−5 and 3×10−6 mean squared position error and 7% and 17% mean percentage force error, for excavation and wheel data, respectively. More importantly, PCA significantly enhances the time and memory efficiency of GNS in both training and rollout. This enables GNS to be trained using a single desktop graphics processing unit (GPU) with moderate video random-access memory (VRAM). This also makes the GNS real-time on large-scale 3D physics configurations (700x faster than our continuum method), which is the first demonstration of such significantly accelerated high-fidelity granular flow engineering simulations.

Opensource machine learning metamodels for assessing blade performance impairment due to general leading edge degradation

Blades leading edge erosion can significantly reduce annual energy production of wind turbines. Accurate estimates of the resulting blade performance impairment are paramount to predict the resulting energy losses and enable cost-informed decisions on optimal maintenance and operational strategies, maximizing energy production and reducing maintenance costs. Computational Fluid Dynamics (CFD) is a robust approach for predicting the performance losses due to LEE. However, the impact of the damage on blade aerodynamics varies depending on damage pattern, extent and location. Therefore, direct CFD simulation of a sufficiently general set of damaged blades is computationally not viable in industrial applications, since the energy loss assessment needs to be performed for hundreds of turbines at many times of the wind farm operation. To address this issue, previous studies showed how CFD can be used to train machine learning metamodels of the perfomance of damaged blade sections, enabling the definition of multi-fidelity energy loss prediction systems. This study presents improved metamodels, using validated CFD to generate training datasets that cover a more general and wider range of erosion patterns, from low-amplitude roughness to severe grooves. In order to provide the industry with additional erosion geometry-linked tools for estimating energy yield losses, and foster further research and development in this area, the developed meta-models have been made available online with unrestricted access.

A time-saving fault diagnosis using simplified fast GAN and triple-type data transfer learning

Existing intelligent fault diagnosis approaches demand substantial data for training diagnostic models. However, factors such as the inherent characteristics of bearings, operating conditions, and privacy security make collecting comprehensive fault-bearing data very difficult. Although generating synthetic data through generative adversarial networks (GANs) is feasible, the data generation of GANs is a time-consuming process. To address these challenges, a fault diagnosis framework based on GAN and deep transfer learning (DTL) is proposed, termed the simplified fast GAN triple-type data transfer learning (SFGAN-TDTL) method. Initially, an SFGAN is proposed as a replacement for traditional GANs. The time-frequency image data generated by SFGAN serve to augment the training dataset, offering faster and higher-quality data generation compared to traditional GANs. To further reduce time consumption for GAN-based methods, the TDTL method is proposed. Differing from DTL, which utilizes synthetic data to construct a pre-trained model and conducts targeted fine-tuning with real data, TDTL employs open-source data, synthetic data, and real data to fill the weights of the task-insensitive layer, task-sensitive layer, and fully connected layer, respectively. Numerical results demonstrate that SFGAN-TDTL maintains higher diagnostic accuracy while significantly reducing time consumption.

Supplementation of deep neural networks with simplified physics-based features to increase accuracy of plate fundamental frequency predictions

To improve predictive machine learning-based models limited by sparse data, supplemental physics-related features are introduced into a deep neural network (DNN). While some approaches inject physics through differential equations or numerical simulation, improvements are possible using simplified relationships from engineering references. To evaluate this hypothesis, thin rectangular plates were simulated to generate training datasets. With plate dimensions and material properties as input features and fundamental natural frequency as the output, predictive performance of a data-driven DNN-based model is compared with models using supplemental inputs, such as modulus of rigidity. To evaluate model accuracy improvements, these additional features are injected into various DNN layers, and the network is trained with four different dataset sizes. When evaluated against independent data of similar features to the training sets, supplementation provides no statistically-significant prediction error reduction. However, notable accuracy gains occur when independent test data is of material and dimensions different from the original training set. Furthermore, when physics-enhanced data is injected into multiple DNN layers, reductions in mean error from 33.2% to 19.6%, 34.9% to 19.9%, 35.8% to 22.4%, and 43.0% to 28.4% are achieved for dataset sizes of 261, 117, 60, and 30, respectively, demonstrating potential for generalizability using a data supplementation approach. Additionally, when compared with other methods—such as linear regression and support vector machine (SVM) approaches—the physics-enhanced DNN demonstrates an order of magnitude reduction in percentage error for dataset sizes of 261, 117, and 60 and a 30% reduction for a size of 30 when compared with a cubic SVM model independently tested with data divergent from the training and validation set.

A probabilistic neural twin for treatment planning in peripheral pulmonary artery stenosis.

The substantial computational cost of high-fidelity models in numerical hemodynamics has, so far, relegated their use mainly to offline treatment planning. New breakthroughs in data-driven architectures and optimization techniques for fast surrogate modeling provide an exciting opportunity to overcome these limitations, enabling the use of such technology for time-critical decisions. We discuss an application to the repair of multiple stenosis in peripheral pulmonary artery disease through either transcatheter pulmonary artery rehabilitation or surgery, where it is of interest to achieve desired pressures and flows at specific locations in the pulmonary artery tree, while minimizing the risk for the patient. Since different degrees of success can be achieved in practice during treatment, we formulate the problem in probability, and solve it through a sample-based approach. We propose a new offline-online pipeline for probabilistic real-time treatment planning which combines offline assimilation of boundary conditions, model reduction, and training dataset generation with online estimation of marginal probabilities, possibly conditioned on the degree of augmentation observed in already repaired lesions. Moreover, we propose a new approach for the parametrization of arbitrarily shaped vascular repairs through iterative corrections of a zero-dimensional approximant. We demonstrate this pipeline for a diseased model of the pulmonary artery tree available through the Vascular Model Repository.

Bridging the gap between high-level quantum chemical methods and deep learning models

Supervised deep learning (DL) models are becoming ubiquitous in computational chemistry because they can efficiently learn complex input-output relationships and predict chemical properties at a cost significantly lower than methods based on quantum mechanics. The central challenge in many DL applications is the need to invest considerable computational resources in generating large ( N>1×105 ) training sets such that the resulting DL model can be generalized reliably to unseen systems. The lack of better alternatives has encouraged the use of low-cost and relatively inaccurate density-functional theory (DFT) methods to generate training data, leading to DL models that lack accuracy and reliability. In this article, we describe a robust and easily implemented approach based on property-specific atom-centered potentials (ACPs) that resolves this central challenge in DL model development. ACPs are one-electron potentials that are applied in combination with a computationally inexpensive but inaccurate quantum mechanical method (e.g. double-ζ DFT) and fitted against relatively few high-level data ( N≈1×103 – 1×104 ), possibly obtained from the literature. The resulting ACP-corrected methods retain the low cost of the double-ζ DFT approach, while generating high-level-quality data in unseen systems for the specific property for which they were designed. With this approach, we demonstrate that ACPs can be used as an intermediate method between high-level approaches and DL model development, enabling the calculation of large and accurate DL training sets for the chemical property of interest. We demonstrate the effectiveness of the proposed approach by predicting bond dissociation enthalpies, reaction barrier heights, and reaction energies with chemical accuracy at a computational cost lower than the DFT methods routinely used for DL training data set generation.

Short-term voltage instability prediction using pre-identified voltage templates and machine learning classifiers

System operators use different procedures to detect disturbances which can lead to short-term voltage instability. These schemes are designed using the domain knowledge and generally specific for the corresponding system. The design procedure of such schemes is heuristic and cannot be directly applied to any other power system. This paper proposes a data-driven approach to predict short-term voltage stability using set of post-disturbance voltage templates and a machine learning-based classifier. In this scheme, a set of voltage templates corresponding to stable/unstable events is identified for selected buses using a suitably generated training data set. The proximity of real-time voltage trajectory to each of the template is computed and input to a trained machine learning classifier to predict the short-term voltage stability status. The proposed approach is validated using IEEE Nordic test system. The investigations show that this method provides accurate short-term voltage stability predictions. Furthermore, impact of adverse phenomena such as measurement noise and missing data on the proposed method is analyzed to confirm the robustness of this scheme in particle environments. © 2017 Elsevier Inc. All rights reserved.

Solving a Class of Cut-Generating Linear Programs via Machine Learning

Cut-generating linear programs (CGLPs) play a key role as a separation oracle to produce valid inequalities for the feasible region of mixed-integer programs. When incorporated inside branch-and-bound, the cutting planes obtained from CGLPs help to tighten relaxations and improve dual bounds. However, running the CGLPs at the nodes of the branch-and-bound tree is computationally cumbersome due to the large number of node candidates and the lack of a priori knowledge on which nodes admit useful cutting planes. As a result, CGLPs are often avoided at default settings of branch-and-cut algorithms despite their potential impact on improving dual bounds. In this paper, we propose a novel framework based on machine learning to approximate the optimal value of a CGLP class that determines whether a cutting plane can be generated at a node of the branch-and-bound tree. Translating the CGLP as an indicator function of the objective function vector, we show that it can be approximated through conventional data classification techniques. We provide a systematic procedure to efficiently generate training data sets for the corresponding classification problem based on the CGLP structure. We conduct computational experiments on benchmark instances using classification methods such as logistic regression. These results suggest that the approximate CGLP obtained from classification can improve the solution time compared with that of conventional cutting plane methods. Our proposed framework can be efficiently applied to a large number of nodes in the branch-and-bound tree to identify the best candidates for adding a cut. History: Accepted by Andrea Lodi, Area Editor/Design & Analysis of Algorithms — Discrete. Supplemental Material: The e-companion is available at https://doi.org/10.1287/ijoc.2022.0241 .

Residential Building Design and Optimization in Arid Climates using Multiple Objectives and ANN-GA

Residential buildings in arid regions, such as South Algeria, are responsible for a considerable portion of the total electrical energy consumption, surpassing 60%. To tackle this challenge, this research proposes a novel multi-objective approach aimed at designing typical residential buildings. This approach combines the utilization of a Genetic Algorithm (GA) optimization technique with an Artificial Neural Network (ANN) model, implemented using the Python programming language and integrated with TRNSYS software. The performance of the ANN model is thoroughly evaluated and validated using the TRNSYS software. The integration of Python and TRNSYS facilitates the generation of training datasets for the ANN model. This co-simulation framework provides a powerful tool for accurately estimating energy consumption and life cycle costs (LCC). The results obtained through the TOPSIS-based analysis reveal a remarkable reduction in cooling and heating energy usage, up to 51.43% and 98.97%, respectively, when compared to the basic scenario.

MTSTR: Multi-task learning for low-resolution scene text recognition via dual attention mechanism and its application in logistics industry.

Recognizing texts in images plays an important role in many applications, such as industrial intelligence, robot vision, automatic driving, command assistance, and scene understanding. Although great progress has been achieved in various fields, research on complex systems modeling using text recognition technology requires further attention. To address this, we propose a new end-to-end multi-task learning method, which includes a super-resolution branch (SRB) and a recognition branch. To effectively learn the semantic information of images, we utilize the feature pyramid network (FPN) to fuse high- and low-level semantic information. The feature map generated by FPN is then delivered separately to the super-resolution branch and the recognition branch. We introduce a novel super-resolution branch, the SRB based on the proposed dual attention mechanism (DAM), designed to enhance the capability of learning low-resolution text features. The DAM incorporates the residual channel attention to enhance channel dependencies and the character attention module to focus on context information. For the recognition branch, the feature map generated by FPN is fed into an RNN sequence module, and an attention-based decoder is constructed to predict the results. To address the issue of low-resolution text recognition in numerous Chinese scenes, we propose the Chinese super-resolution datasets instead of relying on traditional down-sampling techniques to generate training datasets. Experiments demonstrate that the proposed method performs robustly on low-resolution text images and achieves competitive results on benchmark datasets.

An enhanced Sample-Partitioning Adaptive Reduced Chemistry method with a-priori error estimation

Reactor-based approaches for handling the Turbulence–Chemistry-Interactions closure have the advantage of embedding finite-rate chemistry in the combustion model of RANS and LES simulations, which might be crucial for the solution accuracy when complex combustion regimes are investigated. However, the numerical solution of the chemical ODEs is burdened with stiffness and increased dimensionality, especially when large detailed mechanisms are required. To this end, the Sample-Partitioning Adaptive Reduced Chemistry (SPARC) methodology couples adaptive chemistry and machine learning to speed-up the chemistry integration in reactive flows simulations. It consists in building a library of skeletal mechanisms, associated to clusters of similar thermo-chemical states identified in a training dataset, and then, at run-time, each computational cell is assigned to a specific cluster, whose skeletal mechanism is retrieved and employed for the time integration. Such workflow builds on four interacting blocks, i.e., training dataset generation, clustering, mechanism simplification, and classification, and its success tightly depends on the quality of each block, which generally results from a combination of theoretical, methodological, and computational choices. In this paper, we explore the effects of the mechanism simplification strategy on the SPARC performance and we develop an ad-hoc procedure that automatically identifies the cluster-wise optimal reduction parameters, delivering a higher global reduction and therefore a larger computational speed-up compared to a standard approach, along with an explicit a-priori control on accuracy. We implement and test this novel procedure on a RANS simulation of the Adelaide Jet-in-hot-coflow (AJHC) burner, and we attain a ∼2x CPU time improvement with respect to the simulation obtained with a 36-species detailed mechanism.Novelty and significance This work makes a contribution towards the acceleration of chemistry integration in reactive flows simulations. More specifically, the Sample-Partitioning Adaptive Reduced Chemistry method, which couples adaptive chemistry and machine learning, is enhanced with automatic target species definition and a-priori error estimation. The novelties lie in the utilization of the computational singular perturbation (CSP) reduction algorithm, which provides means for automatically identifying an adaptive set of target species, and in the definition of a novel strategy for assessing the performance of the reduced mechanisms in the pre-processing phase, with the goal of estimating a measure of accuracy of the upcoming CFD simulation.

Detection of Low-Velocity Layer Using Passive Surface-Wave Data and Deep Learning: An Application in The Urban Area of Hangzhou, China

Passive surface-wave methods using dense seismic arrays have gained growing attention in near-surface high-resolution imaging in urban environments. Deep learning (DL) can release a tremendous workload brought by dense seismic arrays. We presented a case study of shear-wave velocity (Vs) structure imaging in the Hangzhou urban area (eastern China) using DL inversion. Noise data were recorded by dense linear arrays with approximately 5 m spacing deployed along two crossing roads for investigating the top 80 m of the subsurface. Phase-velocity dispersion curves are extracted from virtual shot gathers using multichannel analysis of surface waves. We divided the area where the low-velocity layer (LVL) may exist into three layers with a thickness of 5 m. We gave the four layers weak constraints to generate training dataset and adopted a convolutional neural network to directly invert fundamental-mode Rayleigh-wave phase velocity for 1D Vs models. To improve the accuracy, we further applied the sensitivities to weight the loss function in DL inversion. The obtained pseudo-2D Vs profiles correspond to the velocities estimated from logging data and previous survey. The well-trained neural network successfully identified that the LVL is located at 50-60 m deep. And this network was also achieved accurately the inversion of a dense seismic network nearby. The results of this survey demonstrate the accuracy and efficiency of delineating near-surface structures from traffic-induced noise using the DL technique, which has great potential for monitoring subsurface changes in urban areas.

A generative model to simulate spatiotemporal dynamics of biomolecules in cells.

Generators of space-time dynamics in bioimaging have become essential to build ground truth datasets for image processing algorithm evaluation such as biomolecule detectors and trackers, as well as to generate training datasets for deep learning algorithms. In this contribution, we leverage a stochastic model, called birth-death-move (BDM) point process, in order to generate joint dynamics of biomolecules in cells. This particle-based stochastic simulation method is very flexible and can be seen as a generalization of well-established standard particle-based generators. In comparison, our approach allows us: (1) to model a system of particles in motion, possibly in interaction, that can each possibly switch from a motion regime (e.g., Brownian) to another (e.g., a directed motion); (2) to take into account finely the appearance over time of new trajectories and their disappearance, these events possibly depending on the cell regions but also on the current spatial configuration of all existing particles. This flexibility enables to generate more realistic dynamics than standard particle-based simulation procedures, by for example accounting for the colocalization phenomena often observed between intracellular vesicles. We explain how to specify all characteristics of a BDM model, with many practical examples that are relevant for bioimaging applications. As an illustration, based on real fluorescence microscopy datasets, we finally calibrate our model to mimic the joint dynamics of Langerin and Rab11 proteins near the plasma membrane, including the well-known colocalization occurrence between these two types of vesicles. We show that the resulting synthetic sequences exhibit comparable features as those observed in real microscopy image sequences.

AMBY: A development environment for youth to create conversational agents

Conversational AIs such as Alexa and ChatGPT are increasingly ubiquitous in young people’s lives, but these young users are often not afforded the opportunity to learn about the inner workings of these technologies. One of the most powerful ways to foster this learning is to empower youth to create AI that is personally and socially meaningful to them. We have built a novel development environment, AMBY–“AI Made By You”–for youth to create conversational agents. AMBY was iteratively designed with and for youth aged 12–13 through contextual inquiry and usability studies. AMBY is designed to foster AI learning with features that enable users to generate training datasets and visualize conversational flow. We report on results from a two-week summer camp deployment, and contribute design implications for conversational AI authoring tools that empower AI learning for youth.