Apply Dimensionality Reduction Research Articles

A Multistage Nonlinear Method for Aeroengine Health Parameter Estimation Based on Adjacent Operating Points

Abstract Health parameter estimation is the core of engine gas path analysis (GPA), which is widely adopted for engine safety improvement, as well as for operation and maintenance cost reduction. The major challenge of GPA lies in the contradiction between the high dimensions of parameters under estimation, e.g., health parameters, and the limited measurements obtainable from a small number of sensors. Existent GPA methods for health parameters commonly apply dimension reduction before estimation, leading to information loss and hence inaccurate estimation. To tackle the challenge of limited sensor measurements and to have more system outputs than parameters under estimation, we proposed to augment the output vector of the system model by combining the measurements from multiple adjacent operating points. But the engine model can face the problem of homogenization if using data from adjacent operating points. This can, in turn, lead to a low identifiability of parameters. We analyze the internal mechanism of such large deviation of the parameter estimation results based on linear models and argue for the need of nonlinear method. Hence, we propose a multistage nonlinear parameter estimation method for health parameters, combining biased and unbiased estimation. In our extensive simulations based on 10 output measurements of a JT9D engine, our method can estimate 130% more parameters than the widely used GPA method, while reducing the maximum estimation error of health parameters from 2.2% to 0.1%.

A computational topology-based spatiotemporal analysis technique for honeybee aggregation

A primary challenge in understanding collective behavior is characterizing the spatiotemporal dynamics of the group. We employ topological data analysis to explore the structure of honeybee aggregations that form during trophallaxis, which is the direct exchange of food among nestmates. From the positions of individual bees, we build topological summaries called CROCKER matrices to track the morphology of the group as a function of scale and time. Each column of a CROCKER matrix records the number of topological features, such as the number of components or holes, that exist in the data for a range of analysis scales, at a given point in time. To detect important changes in the morphology of the group from this information, we first apply dimensionality reduction techniques to these matrices and then use classic clustering and change-point detection algorithms on the resulting scalar data. A test of this methodology on synthetic data from an agent-based model of honeybees and their trophallaxis behavior shows two distinct phases: a dispersed phase that occurs before food is introduced, followed by a food-exchange phase during which aggregations form. We then move to laboratory data, successfully detecting the same two phases across multiple experiments. Interestingly, our method reveals an additional phase change towards the end of the experiments, suggesting the possibility of another dispersed phase that follows the food-exchange phase.

A New Dynamics Analysis Model for Five-Axis Machining of Curved Surface Based on Dimension Reduction and Mapping

The equipment used in various fields contains an increasing number of parts with curved surfaces of increasing size. Five-axis computer numerical control (CNC) milling is the main parts machining method, while dynamics analysis has always been a research hotspot. The cutting conditions determined by the cutter axis, tool path, and workpiece geometry are complex and changeable, which has made dynamics research a major challenge. For this reason, this paper introduces the innovative idea of applying dimension reduction and mapping to the five-axis machining of curved surfaces, and proposes an efficient dynamics analysis model. To simplify the research object, the cutter position points along the tool path were discretized into inclined plane five-axis machining. The cutter dip angle and feed deflection angle were used to define the spatial position relationship in five-axis machining. These were then taken as the new base variables to construct an abstract two-dimensional space and establish the mapping relationship between the cutter position point and space point sets to further simplify the dimensions of the research object. Based on the in-cut cutting edge solved by the space limitation method, the dynamics of the inclined plane five-axis machining unit were studied, and the results were uniformly stored in the abstract space to produce a database. Finally, the prediction of the milling force and vibration state along the tool path became a data extraction process that significantly improved efficiency. Two experiments were also conducted which proved the accuracy and efficiency of the proposed dynamics analysis model. This study has great potential for the online synchronization of intelligent machining of large surfaces.

Service failure monitoring via multivariate multiple linear regression profile schemes with dimensionality reduction

Service failures can be manifested through customer complaints in the form of negative reviews. It is vital for businesses to monitor the occurrence and severity of customer complaints to help balance service failure responses and costs. Existing monitoring schemes can detect complaint anomalies but fail to explain the reasons for these anomalies. We propose new multivariate multiple linear regression profile (MMLRP) based schemes to monitor the relationship between customer complaints and identified explanatory variables. Conventional MMLRP-based schemes suffer performance loss with more than five explanatory variables, and the fixed parameters drawn from a historical process are unsuitable for a changing review-generation process. As such, we apply dimensionality reduction techniques to the explanatory variables and incorporate the variability in new review samples. We use both simulation analysis and an airline service case study to show the effectiveness of our proposed MMLRP-based schemes in customer complaint anomaly detection and diagnosis.

An efficient joint framework assisted by embedded feature smoother and sparse skip connection for hyperspectral image classification

Deep neural network (DNN) methods play an essential role in hyperspectral classification. However, the massive parameters and vast computing overhead of DNN needs to be reduced when facing the deployment with limited storage and computing resources for real-time response applications, especially considering the high dimensionality of hyperspectral image. So, applying dimension reduction (DR) methods is a crucial pre-processing method in various studies. Still, most of them ignore the feature restoration after the data transformation by DR. In neural networks, many works still involve sophisticated skip connections and dense feature reuse, which can lead to feature redundancy and increase computational complexity, especially when DR methods have been applied first. Motivated by these issues, an efficient joint framework assisted by embedded feature smoother (FS) and sparse skip connection (SSC) is proposed in this article. Instead of directly feeding DR data into the subsequent network, we embedded a computing-cheap FS based on isotropic total variation to restore and enhance the spatial features. Furthermore, we proposed a SSC 3D convolution neural network to complete spatial–spectral feature representation and classification. The SSC is embodied in the design of log2n-skip connection to concatenate feature maps instead of dense connection, pruning the number of channels and reducing the model parameters. Experimental results show that the embedded FS significantly improves classification accuracy and is superior to other processing methods. Our framework offers much superior to other state-of-the-art deep learning-based methods considering classification performance and lightweight aspects, especially when using few training samples. Moreover, considering detailed processing steps, our framework has a competitively cheaper time consumption.

Assessing the impact of bag‐of‐words versus word‐to‐vector embedding methods and dimension reduction on anomaly detection from log files

AbstractIn terms of cyber security, log files represent a rich source of information regarding the state of a computer service/system. Automating the process of summarizing log file content represents an important aid for decision‐making, especially given the 24/7 nature of network/service operations. We perform benchmarking over eight distinct log files in order to assess the impact of the following: (1) different embedding methods for developing semantic descriptions of the original log files, (2) applying dimension reduction to the high‐dimensional semantic space, and (3) assessing the impact of using different unsupervised learning algorithms for providing a visual summary of the service state. Benchmarking demonstrates that (1) word‐to‐vector embeddings identified by bidirectional encoder representation from transformers (BERT) without “fine‐tuning” are sufficient to match the performance of Bag‐or‐Words embeddings provided by term frequency‐inverse document frequency (TF‐IDF) and (2) the self‐organizing map without dimension reduction provides the most effective anomaly detector.

O-237 Metabolic classification of embryos and oocytes based on hyperspectral imaging and machine learning

Abstract Study question To classify embryos according to their metabolic fingerprint with a robust, non-invasive methodology Summary answer Our Hyperspectral (HS) phasor analysis coupled to artificial intelligence simultaneously detects 6+ metabolites to profile and classify competent embryos and oocytes safely and consistently. What is known already Metabolism plays a key role in oocyte and embryo developmental competence. However, current methods to measure oocyte and embryo metabolism are either invasive, phototoxic, too slow or based on indirect analysis in cumulus cells or culture media. Some intracellular metabolites are autofluorescent and their concentration can be quantified using HS imaging when using the correct light excitation. HS imaging is an optical technique whereby the full emission spectrum is obtained for each pixel of an image. Thus, raw HS data encodes rich metabolic information of an imaged cell as numerous metabolites are excited simultaneously. Study design, size, duration A total of 96 mouse embryos and 115 oocytes were analyzed. In vivo fertilized embryo cohorts included control, glucose-starved, pyruvate/lactate-starved, and glucose/pyruvate/lactate-starved. Oocyte cohorts included oocytes obtained from young female mice analyzed blindly either immediately after collection or after overnight culture (in vitro aged), and from old female mice. After HS analysis, oocytes were inseminated by ICSI and cultured in vitro to correlate their potential to develop up to blastocyst with their metabolic profiles. Participants/materials, setting, methods Samples were imaged using multiphoton illumination near the infra-red to avoid phototoxicity. The HS detection covered the whole visible spectrum allowing simultaneous measurement of 6+ relevant metabolites for the embryo/oocyte biology: NADH (free and bound), FAD+, retinol, retinoic acid, and flavins among others. Afterwards, we applied dimensionality reduction (2D phasor image) from which machine learning algorithms learned to classify the different samples. We used Akaike Information Criterion (AIC) model. Cross validation was performed using 80%/20%. Main results and the role of chance The Area Under the ROC Curve (AUC) achieved by binary classification as well as the average class-accuracy (computed as the average of the confusion matrix diagonal) were calculated in four-class classification in the case of embryos. [EM1] Brightfield images of the test set of embryos were evaluated by 4 embryologists to look for compromised embryos. Human graders reached an AUC of 51% of the samples using brightfield compared to 93.7% of the AI algorithm using HS-Phasor. A similar approach using the HS-Phasor was conducted on the different cohorts of mouse oocytes. It was able to separate the oocyte classes with an average AUC of 96.2% and make a statistical correlation with their blastulation efficiency with a 82.2% AUC. The viability of embryos and oocytes after imaging was comparable to that of non-imaged controls, indicating that the imaging procedure had no effect on the posterior in vitro development of the embryos. Limitations, reasons for caution The cohort size is limited although statistically robust. Blastulation efficiency will need to be complemented (ongoing work) with implantation measurements. Wider implications of the findings The HS-phasor method has the potential to become a paradigm shift in IVF embryo selection due to the reliable, non-invasive direct measurement of the oocyte/embryo physiology. The adaptation into clinical practice can be straightforward as it offers an additional layer of information for oocyte/embryo assessment without compromising time nor viability. Trial registration number NA

Distributionally Robust Surrogate Optimal Control for High-Dimensional Systems

This article presents a novel methodology for tractably solving optimal control and offline reinforcement learning (RL) problems for high-dimensional systems. This work is motivated by the ongoing challenges of safety, computation, and optimality in high-dimensional optimal control. We address these key questions with the following approach. First, we identify a sequence-modeling surrogate methodology that takes as input the initial state and a time series of control inputs and outputs an approximation of the objective function and trajectories of constraint functions. Importantly this approach entirely absorbs the individual state transition dynamics. The sole dependence on the initial state means we can apply dimensionality reduction to compress the model input while retaining most of its information. Uncertainty in the surrogate objective will affect the resulting optimality. Critically, however, uncertainty in the surrogate constraint functions will lead to infeasibility, i.e., unsafe actions. When considering offline RL, the most significant modeling errors will be encountered on out-of-distribution (OOD) data. Therefore, we apply Wasserstein ambiguity sets to “robustify” our surrogate modeling approach subject to worst case out-of-sample modeling errors based on the distribution of test data residuals. We demonstrate the efficacy of this combined approach through a case study of safe optimal fast charging of a high-dimensional lithium-ion battery model at low temperatures.

Classification of stochastic processes with topological data analysis

SummaryIn this study, we demonstrate that engineered topological features can distinguish time series sampled from different stochastic processes with different noise characteristics, in both balanced and unbalanced sampling schemes. We compare our classification results against the results of the same classification on features coming from descriptive statistics and the wavelet transform. We conclude that machine learning models built on engineered topological features alone perform consistently better than those built on the standard statistical and wavelet features for time series classification tasks. We also apply dimension reduction techniques to our engineered features and compare the result of our classification models before and after dimensionality reduction. Finally, we also show that in our calculations of the engineered topological features, employing parallel computing methods does yield significant improvements in run time and memory footprint.

Quantification of spatial and phenotypic heterogeneity in an agent-based model of tumour-macrophage interactions.

We introduce a new spatial statistic, the weighted pair correlation function (wPCF). The wPCF extends the existing pair correlation function (PCF) and cross-PCF to describe spatial relationships between points marked with combinations of discrete and continuous labels. We validate its use through application to a new agent-based model (ABM) which simulates interactions between macrophages and tumour cells. These interactions are influenced by the spatial positions of the cells and by macrophage phenotype, a continuous variable that ranges from anti-tumour to pro-tumour. By varying model parameters that regulate macrophage phenotype, we show that the ABM exhibits behaviours which resemble the 'three Es of cancer immunoediting': Equilibrium, Escape, and Elimination. We use the wPCF to analyse synthetic images generated by the ABM. We show that the wPCF generates a 'human readable' statistical summary of where macrophages with different phenotypes are located relative to both blood vessels and tumour cells. We also define a distinct 'PCF signature' that characterises each of the three Es of immunoediting, by combining wPCF measurements with the cross-PCF describing interactions between vessels and tumour cells. By applying dimension reduction techniques to this signature, we identify its key features and train a support vector machine classifier to distinguish between simulation outputs based on their PCF signature. This proof-of-concept study shows how multiple spatial statistics can be combined to analyse the complex spatial features that the ABM generates, and to partition them into interpretable groups. The intricate spatial features produced by the ABM are similar to those generated by state-of-the-art multiplex imaging techniques which distinguish the spatial distribution and intensity of multiple biomarkers in biological tissue regions. Applying methods such as the wPCF to multiplex imaging data would exploit the continuous variation in biomarker intensities and generate more detailed characterisation of the spatial and phenotypic heterogeneity in tissue samples.

Improving Accuracy of Credit Card Fraud Detection Using Supervised Machine Learning Models and Dimension Reduction

Credit card fraud is a serious crime, and it is a common type of identity theft. Financial institutions and consumers are experiencing economical losses due to financial fraud caused by credit card transactions. Machine Learning Models can aid and alleviate credit card fraud by providing real time detection of credit card fraud before it takes place. The problem that arises with machine learning models is poor performance in terms of accuracy if the data objects in dataset have high dimensionality. In this paper we have tested and compared six machine learning models in detecting credit card fraud. Furthermore, dimension reduction techniques was used to improve the performance of these machine learning models. The results show improved accuracy on the machine learning models after applying dimension reduction and removing anomalies and imbalance.

Spatial-spectral classification of hyperspectral remote sensing images using 3D CNN based LeNet-5 architecture

Hyperspectral remote sensing image (HRSI) analysis are commonly used in a wide variety of remote sensing applications such as land cover analysis, military surveillance, object detection and precision agriculture. Deep learning is seen as an effective method to improve HRSI classification. In particular, convolutional neural networks (CNNs) are increasingly used in this field. The high dimensionality of the HRSIs increases the computational complexity. Thus, most of studies apply dimension reduction as preprocessing. Another problem in HRSI classification is that spatial-spectral features must be considered in order to obtain accurate results. Because, HRSI classification results are highly dependent on spatiospectral information. The aim of this paper is to build a 3D CNN-based LeNet-5 method for HRSI classification. Principal component analysis (PCA) is used as the pre-processing step for optimum spectral band extraction. 3D CNN is used to simultaneously extract spatial-spectral features in HRSIs. LeNet-5 architecture has a simple and straightforward architecture. At the same time, the number of trainable parameters is very low. With the use of the LeNet-5 architecture, the number of trainable parameters of the proposed method is considerably reduced. This is one of the most important features that distinguish the proposed method from other deep learning methods. The proposed method is tested with Indian pines, Pavia University and Salinas datasets. As a result of experimental studies, 100% overall accuracy result is obtained in all datasets. The proposed 3DLeNet method is compared against various state-of-the-art CNN based methods. From the experimental results, it is seen that our 3DLeNet method performs more accurate result. It has also been found that the proposed 3DLeNet method shows a satisfactory result with less computational complexity.

Global and regional changes in carbon dioxide emissions: 1970–2019

We introduce new frameworks to study spatio-temporal patterns in carbon dioxide emissions, demographic trends and economic patterns across 50 countries over the past 50 years. Our analysis is broken up into four sections. First, we introduce a new method to classify countries into one of three characteristic emissions classes based on a one, two or three-segment piecewise linear model. We reveal that most countries are best represented by a piecewise linear model with one change point. Next, we perform a decade-by-decade study of carbon dioxide trajectories. There, we demonstrate notable changes in cluster structures in each decade. We then study the spatial propagation of emissions over time, highlighting a peak in spatial dispersion in 2000, beyond which there has been a gradual decline in spatial emissions variance over space. Finally, we use carbon dioxide, GDP, and population data and apply dimensionality reduction and clustering to group countries based on similarity in their real and carbon economies.

Immune Subtypes and Immune Landscape Analysis of Endometrial Carcinoma.

Some patients with endometrial cancer (EC) suffer from limited survival benefits after immunotherapy, suggesting that there may be a specific pattern associated with immunotherapy. Immune-related genes were extracted from The Cancer Genome Atlas databases. We analyzed the differences among immune subtypes (ISs) in the distribution of the tumor mutational burden, chemotherapy-induced immune response markers, immune checkpoint-related genes, immunotherapy, and chemotherapy. We applied dimensionality reduction and defined the immune landscape of EC. Then, we used the Weighted Gene Co-Expression Network Analysis package to identify the coexpression modules of these immune genes. Finally, hub genes were selected and detected by quantitative PCR and immunohistochemistry. We obtained three ISs. There were differences in the distribution of the tumor mutational burden, chemotherapy-induced immune response markers, and immune checkpoint-related genes among the ISs. Regarding immunotherapy and chemotherapy, the IS2 subtypes were more sensitive to programmed cell death protein 1 inhibitors. In addition, different positions in the immune landscape map exhibited different prognostic characteristics, providing further evidence of the ISs. The IS2 subtypes were significantly positively correlated with yellow module gene list, indicating a good prognosis with high score. SIRPG and SLAMF1 were identified as the final characteristic genes. The quantitative PCR and immunohistochemistry results showed that the expression levels of SIRPG and SLAMF1 were low in human EC tissue. In this study, we identified three reproducible ISs of EC. The immune landscape analysis further revealed the intraclass heterogeneity of the ISs. SIRPG and SLAMF1 were identified to be associated with progression, suggesting that they may be novel immune-related biomarkers of EC.

Considerations for Assessing Functional Forest Diversity in High-Dimensional Trait Space Derived from Drone-Based Lidar

Remotely sensed morphological traits have been used to assess functional diversity of forests. This approach is potentially spatial-scale-independent. Lidar data collected from the ground or by drone at a high point density provide an opportunity to consider multiple ecologically meaningful traits at fine-scale ecological units such as individual trees. However, high-spatial-resolution and multi-trait datasets used to calculate functional diversity can produce large volumes of data that can be computationally resource demanding. Functional diversity can be derived through a trait probability density (TPD) approach. Computing TPD in a high-dimensional trait space is computationally intensive. Reductions of the number of dimensions through trait selection and principal component analysis (PCA) may reduce the computational load. Trait selection can facilitate identification of ecologically meaningful traits and reduce inter-trait correlation. This study investigates whether kernel density estimator (KDE) or one-class support vector machine (SVM) may be computationally more efficient in calculating TPD. Four traits were selected for input into the TPD: canopy height, effective number of layers, plant to ground ratio, and box dimensions. When simulating a high-dimensional trait space, we found that TPD derived from KDE was more efficient than using SVM when the number of input traits was high. For five or more traits, applying dimension reduction techniques (e.g., PCA) are recommended. Furthermore, the kernel size for TPD needs to be appropriate for the ecological target unit and should be appropriate for the number of traits. The kernel size determines the required number of data points within the trait space. Therefore, 3–5 traits require a kernel size of at least 7×7pixels. This study contributes to improving the quality of TPD calculations based on traits derived from remote sensing data. We provide a set of recommendations based on our findings. This has the potential to improve reliability in identifying biodiversity hotspots.

A dynamic gradient architecture generates brain activity states

The human brain exhibits a diverse yet constrained range of activity states. While these states can be faithfully represented in a low-dimensional latent space, our understanding of the constitutive functional anatomy is still evolving. Here we applied dimensionality reduction to task-free and task fMRI data to address whether latent dimensions reflect intrinsic systems and if so, how these systems may interact to generate different activity states. We find that each dimension represents a dynamic activity gradient, including a primary unipolar sensory-association gradient underlying the global signal. The gradients appear stable across individuals and cognitive states, while recapitulating key functional connectivity properties including anticorrelation, modularity, and regional hubness. We then use dynamical systems modeling to show that gradients causally interact via state-specific coupling parameters to create distinct brain activity patterns. Together, these findings indicate that a set of dynamic, intrinsic spatial gradients interact to determine the repertoire of possible brain activity states.

Detection of Covid-19 Using CAD System Depending on Chest X-Ray and Machine Learning Techniques

SARS-COV-2 (severe acute respiratory syndrome coronavirus-2) has caused widespread mortality. Infected individuals had specific radiographic visual features and fever, dry cough, lethargy, dyspnea, and other symptoms. According to the study, the chest X-ray (CXR) is one of the essential non-invasive clinical adjuncts for detecting such visual reactions associated with SARS-COV-2. Manual diagnosis is hindered by a lack of radiologists’ availability to interpret CXR images and by the faint appearance of illness radiographic responses. The paper describes an automatic COVID detection based on the deep learning-based system that applied transfer learning techniques to extract features from CXR images to distinguish. The system has three main components. The first part is extracting CXR features with MobileNetV2. The second part used the extracted features and applied Dimensionality reduction using LDA. The final part is a Classifier, which employed XGBoost to classify dataset images into Normal, Pneumonia, and Covid-19. The proposed system achieved both immediate and high results with an overall accuracy of 0.96%, precision of 0.95%, recall of 0.94%, and F1 score of 0.94%.

SurfNet: Learning Surface Representations via Graph Convolutional Network

AbstractFor scientific visualization applications, understanding the structure of a single surface (e.g., stream surface, isosurface) and selecting representative surfaces play a crucial role. In response, we propose SurfNet, a graph‐based deep learning approach for representing a surface locally at the node level and globally at the surface level. By treating surfaces as graphs, we leverage a graph convolutional network to learn node embedding on a surface. To make the learned embedding effective, we consider various pieces of information (e.g., position, normal, velocity) for network input and investigate multiple losses. Furthermore, we apply dimensionality reduction to transform the learned embeddings into 2D space for understanding and exploration. To demonstrate the effectiveness of SurfNet, we evaluate the embeddings in node clustering (node‐level) and surface selection (surface‐level) tasks. We compare SurfNet against state‐of‐the‐art node embedding approaches and surface selection methods. We also demonstrate the superiority of SurfNet by comparing it against a spectral‐based mesh segmentation approach. The results show that SurfNet can learn better representations at the node and surface levels with less training time and fewer training samples while generating comparable or better clustering and selection results.

White matter microstructural variability linked to differential attentional skills and impulsive behavior in a pediatric population.

Structural and functional magnetic resonance imaging (MRI) studies have suggested a neuroanatomical basis that may underly attention-deficit-hyperactivity disorder (ADHD), but the anatomical ground truth remains unknown. In addition, the role of the white matter (WM) microstructure related to attention and impulsivity in a general pediatric population is still not well understood. Using a state-of-the-art structural connectivity pipeline based on the Brainnetome atlas extracting WM connections and its subsections, we applied dimensionality reduction techniques to obtain biologically interpretable WM measures. We selected the top 10 connections-of-interests (located in frontal, parietal, occipital, and basal ganglia regions) with robust anatomical and statistical criteria. We correlated WM measures with psychometric test metrics (Conner's Continuous Performance Test 3) in 171 children (27 Dx ADHD, 3Dx ASD, 9-13years old) from the population-based GESTation and Environment cohort. We found that children with lower microstructural complexity and lower axonal density show a higher impulsive behavior on these connections. When segmenting each connection in subsections, we report WM alterations localized in one or both endpoints reflecting a specific localization of WM alterations along each connection. These results provide new insight in understanding the neurophysiology of attention and impulsivity in a general population.

Statistical learning for predicting density–matrix‐based electron dynamics

We consider the problem of learning density‐dependent molecular Hamiltonian matrices from time series of electron density matrices, all in the context of Hartree–Fock theory. Prior work developed a solution to this problem for small molecular systems with density and Hamiltonian matrices of size at most 6 × 6. Here, using a battery of techniques, we scale prior methods to larger molecular systems with, for example, 29 × 29 matrices. This includes systems that either have more electrons or are expressed in large basis sets such as 6‐311++G**. Scaling the method to larger systems enhances its relevance for realistic applications in chemistry and physics. To achieve this scaling, we apply dimensionality reduction, ridge regression and analytic computation of Hessians. Through the combination of these techniques, we are able to learn Hamiltonians by minimizing an objective function that encodes local propagation error. Importantly, these learned Hamiltonians can then be used to predict electron dynamics for thousands of steps: When we use our learned Hamiltonians to numerically solve the time‐dependent Hartree–Fock equation, we obtain predicted dynamics that are in close quantitative agreement with ground truth dynamics. This includes field‐off trajectories similar to the training data and field‐on trajectories outside of the training data.