Search NASA⌕ Search

SEARCH · Search NASA

Results for “timing error evaluation”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,045 records · Page 58

Comparative Analysis of Simulated Annealing (SA) and Simplified Generalized SA (SGSA) for Estimation Optimal of Parametric Functional in CATIVIC

Application of simulated annealing (SA) and simplified GSA (SGSA) techniques for parameter optimization of parametric quantum chemistry method (CATIVIC) was performed. A set of organic molecules were selected for test these techniques. Comparison of the algorithms was carried out for error function minimization with respect to experimental values. Results show that SGSA is more efficient than SA with respect to computer time. Accuracy is similar in both methods; however, there are important differences in the final set of parameters.

74 ATOMIC AND MOLECULAR PHYSICS↗

Diverse Applications of Occultation Data in Ozone Assimilation

Ozone profiles from solar occultation instruments provide invaluable information that can be used to evaluate the quality of assimilated ozone fields, from case studies to long-term phenomena. Occultation data can also be applied to include physical constraints while developing components of an assimilation system. Conversely, assimilation of occultation data can help in their validation, and it provides a framework for evaluation of the impact of occultation data on constraining global ozone fields within models. We illustrate these diverse applications by a series of examples using the ozone assimilation system at NASA/Goddard. In a case study, low ozone in the lower stratosphere due to transport of air from the Tropics to northern high latitudes that was captured by assimilation of Aura Microwave Limb Sounder (MLS) and Solar Backscatter Ultraviolet (SBUV) data, was found to agree with SAGE III data. For long-term monitoring, the quality of a multi-year SBUV-only assimilation was evaluated using monthly-mean time series of POAM, HALOE, and SAGE I1 data. We found realism in the representation of the annual cycle in ozone and in some aspects of interannual variability. Assimilation of POAM data was shown to improve the representation of lower stratospheric ozone, especially over Antarctica. More recently, we assimilated ILAS II ozone data in order to help in their validation. Solar occultation data are used to estimate parameters in a new model for forecast error variances that is being developed. These examples demonstrate the importance of occultation data for ozone assimilation, and potential of assimilation to increase the impact and the value of occultation data.

Markowitz, A.↗

Integrated Precision Landing Performance and Technology Assessments of a Human-Scale Mars Lander Using a Generalized Simulation Framework

Human-scale missions to Mars will likely require multiple landers delivered precisely to designated locations. The current NASA human Mars reference architecture assumes delivery of three 25 t payloads from a 1- or 5-Sol orbit to the surface with a landing precision of 50 m to ensure logistics are located near the habitat. While initial navigation estimates improve with on-orbit ground tracking, errors increase during post-deorbit coast. Likewise, Mars atmospheric variability and forecasting uncertainty means that the entry vehicle guidance, navigation, and control systems must be robust to accommodate landing during any time of day or Mars year, including during dust storms. Precision landing technologies are currently being assessed to determine if onboard navigation sensors are sufficient to enable the landing accuracy required or if additional navigation aids such as surface or orbiting beacons will be needed. This study evaluates the system performance requirements to meet the desired landing accuracy for the reference vehicle design and entry, descent, and landing concept of operations. A detailed six degree-of-freedom integrated performance simulation framework is used to perform the assessment and demonstrate that under current assumptions, onboard navigation sensors are sufficient to support precision landing.

spacecraft navigation↗

Integrated Precision Landing Performance and Technology Assessments of a Human- Scale Mars Lander Using a Generalized Simulation Framework

Human-scale missions to Mars will likely require multiple landers delivered precisely to designated locations. The current NASA human Mars reference architecture assumes delivery of three 25 t payloads from a 1- or 5-Sol orbit to the surface with a landing precision of 50 m to ensure logistics are located near the habitat. While initial navigation estimates improve with on-orbit ground tracking, errors increase during post-deorbit coast. Likewise, Mars atmospheric variability and forecasting uncertainty means that the entry vehicle guidance, navigation, and control systems must be robust to accommodate landing during any time of day or Mars year, including during dust storms. Precision landing technologies are currently being assessed to determine if onboard navigation sensors are sufficient to enable the landing accuracy required or if additional navigation aids such as surface or orbiting beacons will be needed. This study evaluates the system performance requirements to meet the desired landing accuracy for the reference vehicle design and entry, descent, and landing concept of operations. A detailed six degree-of-freedom integrated performance simulation framework is used to perform the assessment and demonstrate that under current assumptions, onboard navigation sensors are sufficient to support precision landing.

Chen, Po-Ting↗

An algebraic method for constructing stable and consistent autoregressive filters

In this paper, we introduce an algebraic method to construct stable and consistent univariate autoregressive (AR) models of low order for filtering and predicting nonlinear turbulent signals with memory depth. By stable, we refer to the classical stability condition for the AR model. By consistent, we refer to the classical consistency constraints of Adams–Bashforth methods of order-two. One attractive feature of this algebraic method is that the model parameters can be obtained without directly knowing any training data set as opposed to many standard, regression-based parameterization methods. It takes only long-time average statistics as inputs. The proposed method provides a discretization time step interval which guarantees the existence of stable and consistent AR model and simultaneously produces the parameters for the AR models. In our numerical examples with two chaotic time series with different characteristics of decaying time scales, we find that the proposed AR models produce significantly more accurate short-term predictive skill and comparable filtering skill relative to the linear regression-based AR models. These encouraging results are robust across wide ranges of discretization times, observation times, and observation noise variances. Finally, we also find that the proposed model produces an improved short-time prediction relative to the linear regression-based AR-models in forecasting a data set that characterizes the variability of the Madden–Julian Oscillation, a dominant tropical atmospheric wave pattern.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Time Variable Gravity modeling for Precise Orbits Across the TOPEX/Poseidon, Jason-l and Jason-2 Missions

Modeling of the Time Variable Gravity (TVG) is believed to constitute one of the the largest remaining source of orbit error for altimeter satellite POD. The GSFC operational TVG model consists of forward modeling the atmospheric gravity using ECMWF 6-hour pressure data, a GRACE derived 20x20 annual field to account for changes in the hydrology and ocean water mass, and linear rates for C20, C30, C40, based on 17 years of SLR data analysis (IERS 2003) using the EIGEN-GL04S1 (a GRACE+Lageos-based geopotential solution). Although the GSFC Operational model can be applied from 1987, there may be long-term variations not captured by these linear models, and more importantly the linear models may not be consistent with more recent surface mass trends due to global climate change, We have evaluated the impact of TVG in two different wavs: (1) by using the more recent EIGEN-6S gravity model developed by the GFZ/GRGS tearm, which consists of annual, semi-annual and secular changes in the coefficients to 50x50 determined over 8(?) years of GRACE+Lageos+GOCE data (2003-200?): (2) Application of 4x4 solutions developed from a multi satellite SLR+DORIS solution based on GGM03S that span the period from 1993 to 2011. We have evaluated the recently released EIGEN6s static and time-varying gravity field for Jason-2 (J2). Jason-I (J1), and TOPEX/Posiedon (TP) Precise Orbit Determination (POD) spanning 1993-2011. Although EIGEN6s shows significant improvement for J2POD spanning 2008 - 2011, it also shows significant degradation for TP POD from 1992. The GSFC 4x4 time SLR+DORIS-based series spans 1993 to mid 2011, and shows promise for POD. We evaluate the performance of the different TVG models based on analysis of tracking data residuals use of independent data such as altimeter crossovers, and through analysis of differences with internally-generated and externally generated orbits.

Zelensky, Nikita P.↗

Pan-tropical daily L-band microwave land surface emissivity retrieval from GNSS-R observations

The uncertainties in microwave land surface emissivity (MLSE) measurements have long limited the use of spaceborne microwave radiometer data. As an emerging observation method, Global Navigation Satellite System Reflectometry (GNSS-R) has demonstrated great potential in several land and ocean applications. In this study, a method for obtaining daily MLSE dataset in the pan-tropical region from Cyclone GNSS (CYGNSS) observations is presented and evaluated. The CYGNSS observations are first aggregated into the Equal-Area-Scalable-Earth (EASE) 2.0 36 km grid by a combined weight function of distance, time, and signal-to-noise ratio variance. Then, the method employs a pixel-by-pixel regression algorithm to conduct the daily MLSE retrieval using reference emissivity derived from the Soil Moisture Active Passive (SMAP) brightness temperature. The CYGNSS MLSE shows good agreement with SMAP MLSE, delivering an overall root-mean-square error (RMSE) of 0.022 and 0.017 for horizontal and vertical polarization, respectively, during the training set spanning the whole year of 2018. Furthermore, on the test set from January 2019 to May 2019, the RMSE values amounted to 0.030 and 0.023 for horizontal and vertical polarization, respectively. Temperature records from the International Soil Moisture Network are employed to calculate the emissivity and for in-situ validation, which yield an RMSE of 0.034 and 0.026 for the two polarizations, respectively. The proposed algorithm provides an encouraging approach to obtain accurate daily MLSE dataset for microwave remote sensing. Compared to the SMAP MLSE, the CYGNSS MLSE has a remarkable improvement of 86% in temporal resolution, greatly complementing the existing microwave emissivity datasets.

54 ENVIRONMENTAL SCIENCES↗

Evaluation of an active magnetic resonance tracking system for interstitial brachytherapy

Purpose: In gynecologic cancers, magnetic resonance (MR) imaging is the modality of choice for visualizing tumors and their surroundings because of superior soft-tissue contrast. Real-time MR guidance of catheter placement in interstitial brachytherapy facilitates target coverage, and would be further improved by providing intraprocedural estimates of dosimetric coverage. A major obstacle to intraprocedural dosimetry is the time needed for catheter trajectory reconstruction. Herein the authors evaluate an active MR tracking (MRTR) system which provides rapid catheter tip localization and trajectory reconstruction. The authors assess the reliability and spatial accuracy of the MRTR system in comparison to standard catheter digitization using magnetic resonance imaging (MRI) and CT. Methods: The MRTR system includes a stylet with microcoils mounted on its shaft, which can be inserted into brachytherapy catheters and tracked by a dedicated MRTR sequence. Catheter tip localization errors of the MRTR system and their dependence on catheter locations and orientation inside the MR scanner were quantified with a water phantom. The distances between the tracked tip positions of the MRTR stylet and the predefined ground-truth tip positions were calculated for measurements performed at seven locations and with nine orientations. To evaluate catheter trajectory reconstruction, fifteen brachytherapy catheters were placed into a gel phantom with an embedded catheter fixation framework, with parallel or crossed paths. The MRTR stylet was then inserted sequentially into each catheter. During the removal of the MRTR stylet from within each catheter, a MRTR measurement was performed at 40 Hz to acquire the instantaneous stylet tip position, resulting in a series of three-dimensional (3D) positions along the catheter’s trajectory. A 3D polynomial curve was fit to the tracked positions for each catheter, and equally spaced dwell points were then generated along the curve. High-resolution 3D MRI of the phantom was performed followed by catheter digitization based on the catheter’s imaging artifacts. The catheter trajectory error was characterized in terms of the mean distance between corresponding dwell points in MRTR-generated catheter trajectory and MRI-based catheter digitization. The MRTR-based catheter trajectory reconstruction process was also performed on three gynecologic cancer patients, and then compared with catheter digitization based on MRI and CT. Results: The catheter tip localization error increased as the MRTR stylet moved further off-center and as the stylet’s orientation deviated from the main magnetic field direction. Fifteen catheters’ trajectories were reconstructed by MRTR. Compared with MRI-based digitization, the mean 3D error of MRTR-generated trajectories was 1.5 ± 0.5 mm with an in-plane error of 0.7 ± 0.2 mm and a tip error of 1.7 ± 0.5 mm. MRTR resolved ambiguity in catheter assignment due to crossed catheter paths, which is a common problem in image-based catheter digitization. In the patient studies, the MRTR-generated catheter trajectory was consistent with digitization based on both MRI and CT. Conclusions: The MRTR system provides accurate catheter tip localization and trajectory reconstruction in the MR environment. Relative to the image-based methods, it improves the speed, safety, and reliability of the catheter trajectory reconstruction in interstitial brachytherapy. MRTR may enable in-procedural dosimetric evaluation of implant target coverage.

60 APPLIED LIFE SCIENCES↗

WE-AB-BRA-09: Registration of Preoperative MRI to Intraoperative Radiographs for Automatic Vertebral Target Localization

Purpose: Accurate localization of target vertebrae is essential to safe, effective spine surgery, but wrong-level surgery occurs with surprisingly high frequency. Recent research yielded the “LevelCheck” method for 3D-2D registration of preoperative CT to intraoperative radiographs, providing decision support for level localization. We report a new method (MR-LevelCheck) to perform 3D-2D registration based on preoperative MRI, presenting a solution for the increasingly common scenario in which MRI (not CT) is used for preoperative planning. Methods: Direct extension of LevelCheck is confounded by large mismatch in image intensity between MRI and radiographs. The proposed method overcomes such challenges with a simple vertebrae segmentation. Using seed points at centroids, vertebrae are segmented using continuous max-flow method and dilated by 1.8 mm to include surrounding cortical bone (inconspicuous in T2w-MRI). MRI projections are computed (analogous to DRR) using segmentation and registered to intraoperative radiographs. The method was tested in a retrospective IRB-approved study involving 11 patients undergoing cervical, thoracic, or lumbar spine surgery following preoperative MRI. Registration accuracy was evaluated in terms of projection-distance-error (PDE) between the true and estimated location of vertebrae in each radiograph. Results: The method successfully registered each preoperative MRI to intraoperative radiographs and maintained desirable properties of robustness against image content mismatch, and large capture range. Segmentation achieved Dice coefficient = 89.2 ± 2.3 and mean-absolute-distance (MAD) = 1.5 ± 0.3 mm. Registration demonstrated robust performance under realistic patient variations, with PDE = 4.0 ± 1.9 mm (median ± iqr) and converged with run-time = 23.3 ± 1.7 s. Conclusion: The MR-LevelCheck algorithm provides an important extension to a previously validated decision support tool in spine surgery by extending its utility to preoperative MRI. With initial studies demonstrating PDE <5 mm and 0% failure rate, the method is now in translation to larger scale prospective clinical studies. S. Vogt and G. Kleinszig are employees of Siemens Healthcare.

60 APPLIED LIFE SCIENCES↗

Efficient Training of Deep Neural Operator Networks via Randomized Sampling

Neural operators (NOs) employ deep neural networks to learn the mappings between infinitedimensional function spaces. Deep operator network (DeepONet), a popular NO architecture, has demonstrated success in the real-time prediction of complex dynamics across various scientific and engineering applications. In this work, we introduce a random sampling technique to be adopted during the training of DeepONet, aimed at improving the generalization ability of the model, while significantly reducing the computational time. The proposed approach targets the trunk network of the DeepONet model that outputs the basis functions corresponding to the spatiotemporal locations of the bounded domain on which the physical system is defined. While constructing the loss function, DeepONet training traditionally considers a uniform grid of spatiotemporal points at which all the output functions are evaluated for each iteration. This approach leads to a larger batch size, resulting in poor generalization and increased memory demands, due to the limitations of the stochastic gradient descent (SGD) optimizer. The proposed random sampling over the inputs of the trunk net mitigates these challenges, improving generalization and reducing the memory requirements during training, resulting in significant computational gains. We validate our hypothesis through three benchmark examples, demonstrating substantial reductions in training time while achieving comparable or lower overall test errors relative to the traditional training approach. Our results indicate that incorporating randomization in the trunk network inputs during training enhances the efficiency and robustness of DeepONet, offering a promising avenue for improving the framework’s performance in modeling complex physical systems.

Karumuri, Sharmila [Department of Civil & Systems ↗

LADR: low-cost application-level detector for reducing silent output corruptions

Applications running on future high performance computing (HPC) systems are more likely to experience transient faults due to technology scaling trends with respect to higher circuit density, smaller transistor size and near-threshold voltage (NTV) operations. A transient fault could corrupt application state without warning, possibly leading to incorrect application output. Such errors are called silent data corruptions (SDCs). In this paper, we present LADR, a low-cost application-level SDC detector for scientific applications. LADR protects scientific applications from SDCs by watching for data anomalies in their state variables (those of scientific interest). It employs compile-time data-flow analysis to minimize the number of monitored variables, thereby reducing runtime and memory overheads while maintaining a high level of fault coverage with low false positive rates. We evaluated LADR with 4 scientific workloads and results show that LADR achieved < 80% fault coverage with only ∼ 3% runtime overheads and ∼ 1% memory overheads. As compared to prior state-of-the-art anomaly-based detection methods, SDC achieved comparable or improved fault coverage, but reduced runtime overheads by 21% ∼ 75%, and memory overheads by 35% ∼ 55% for the evaluated workloads. We believe that such an approach with low memory and runtime overheads coupled with attractive detection precision makes LADR a viable approach for assuring the correct output from large-scale high performance simulations.

Chen, Chao↗

Mastering HPC Runtime Prediction: From Observing Patterns to a Methodological Approach

The continual expansion of high-performance computing (HPC) brings with it an increasing need for efficiency. Heavy investment in energy, hardware, and software infrastructure to support peta- and exascale computing requires the optimization of existing systems and, wherever possible, the discernment and adoption of best-practices towards these goals. Such is the case for runtime prediction. When a job is submitted to an HPC system, an estimate of its runtime is provided by the user in the form of "requested wallclock". Error in this user-provided estimate can lead to jobs being prematurely killed by the scheduler, increased wait time on the queue, and decreased system utilization. More than fifteen years of research has been directed at mitigating these effects by using data-driven runtime predictions. Codified here is a set of commonalities and insights emerging from this body of work, which we present as recommendations and best practices. These practices are combined into a methodological approach described and evaluated on an 11-million-job dataset from the National Renewable Energy Laboratory's petascale HPC system, Eagle. This dataset and the accompanying codebase have been released to the public domain for the benefit of the wider HPC research community.

high performance computing↗

Telerobotic control of a dextrous manipulator using master and six-DOF hand-controllers for space assembly and servicing tasks

Two studies were conducted evaluating methods of controlling a telerobot; bilateral force reflecting master controllers and proportional rate six degrees of freedom (DOF) hand controllers. The first study compared the controllers on performance of single manipulator arm tasks, a peg-in-the-hole task, and simulated satellite orbital replacement unit changeout. The second study, a Space Station truss assembly task, required simultaneous operation of both manipulator arms (all 12 DOFs) and complex multiaxis slave arm movements. Task times were significantly longer and fewer errors were committed with the hand controllers. The hand controllers were also rated significantly higher in cognitive and manual control workload on the two-arm task. The master controllers were rated significantly higher in physical workload. There were no significant differences in ratings of manipulator control quality.

O'Hara, John M.↗

Fuzzy logic based robotic controller

Existing Proportional-Integral-Derivative (PID) robotic controllers rely on an inverse kinematic model to convert user-specified cartesian trajectory coordinates to joint variables. These joints experience friction, stiction, and gear backlash effects. Due to lack of proper linearization of these effects, modern control theory based on state space methods cannot provide adequate control for robotic systems. In the presence of loads, the dynamic behavior of robotic systems is complex and nonlinear, especially where mathematical modeling is evaluated for real-time operators. Fuzzy Logic Control is a fast emerging alternative to conventional control systems in situations where it may not be feasible to formulate an analytical model of the complex system. Fuzzy logic techniques track a user-defined trajectory without having the host computer to explicitly solve the nonlinear inverse kinematic equations. The goal is to provide a rule-based approach, which is closer to human reasoning. The approach used expresses end-point error, location of manipulator joints, and proximity to obstacles as fuzzy variables. The resulting decisions are based upon linguistic and non-numerical information. This paper presents a solution to the conventional robot controller which is independent of computationally intensive kinematic equations. Computer simulation results of this approach as obtained from software implementation are also discussed.

Attia, F.↗

An Experimental Evaluation of the Performance of Two Combination Pitot Pressure Probes

Experimental tests have been completed which recorded the ability of two combination steady state and high response time varying Pitot probe designs to accurately measure steady stagnation pressure at a single location in a flow field. Tests were conducted of double-barreled and coannular Prati probes in a 3.5 in. diameter free jet probe calibration facility from Mach 0.1 to 0.9. Geometric symmetry and pitch (-40 deg to 40 deg) and yaw (0 deg to 40 deg) angle actuation were used to fully evaluate the probes. These tests revealed that the double-barreled configuration induced error in its steady state measurement at zero incidence that increased consistently with jet Mach number to 1.1 percent at Mach 0.9. For all Mach numbers, the double-barreled probe nulled at a pitch angle of approximately 7.0 deg and provided inconsistent measurements when yawed. The double-barreled probe provided adequate measurements via both its steady state and high response tubes (within +/- 0.15 percent accuracy) over unacceptable ranges of biased pitch and inconsistent yaw angles which varied with Mach number. By comparison, the coannular probe provided accurate measurements (at zero incidence) for all jet Mach numbers as well as over a flow angularity range which varied from +/- 26.0 deg at Mach 0.3 deg to +/- 14.0 deg at Mach 0.9. Based on these results, the Prati probe is established as the preferred design. Further experimental tests are recommended to document the frequency response characteristics of the Prati probe.

Arend, David J.↗

Assessment of 10 Year Record of Aerosol Optical Depth from OMI UV Observations

The Ozone Monitoring Instrument (OMI) onboard the EOS-Aura satellite provides information on aerosol optical properties by making use of the large sensitivity to aerosol absorption in the near-ultraviolet (UV) spectral region. Another important advantage of using near UV observations for aerosol characterization is the low surface albedo of all terrestrial surfaces in this spectral region that reduces retrieval errors associated with land surface reflectance characterization. In spite of the 13 × 24 square kilometers coarse sensor footprint, the OMI near UV aerosol algorithm (OMAERUV) retrieves aerosol optical depth (AOD) and single-scattering albedo under cloud-free conditions from radiance measurements at 354 and 388 nanometers. We present validation results of OMI AOD against space and time collocated Aerosol Robotic Network measured AOD values over multiple stations representing major aerosol episodes and regimes. OMAERUV's performance is also evaluated with respect to those of the Aqua-MODIS Deep Blue and Terra-MISR AOD algorithms over arid and semi-arid regions in Northern Africa. The outcome of the evaluation analysis indicates that in spite of the "row anomaly" problem, affecting the sensor since mid-2007, the long-term aerosol record shows remarkable sensor stability.

aerosol optical depth↗

Mastering HPC Runtime Prediction: From Observing Patterns to a Methodological Approach: Preprint

The continual expansion of high-performance computing (HPC) brings with it an increasing need for efficiency. Heavy investment in energy, hardware, and software infrastructure to support peta- and exascale computing requires the optimization of existing systems and, wherever possible, the discernment and adoption of best-practices towards these goals. Such is the case for runtime prediction. When a job is submitted to an HPC system, an estimate of its runtime is provided by the user in the form of "requested wallclock''. Error in this user-provided estimate can lead to jobs being prematurely killed by the scheduler, increased wait time on the queue, and decreased system utilization. More than fifteen years of research has been directed at mitigating these effects by using data-driven runtime predictions. Codified here is a set of commonalities and insights emerging from this body of work, which we present as recommendations and best practices. These practices are combined into a methodological approach described and evaluated on an 11-million-job dataset from the National Renewable Energy Laboratory's petascale HPC system, Eagle. This dataset and the accompanying codebase have been released to the public domain for the benefit of the wider HPC research community.

high performance computing↗

Neural Image Compression: Generalization, Robustness, and Spectral Biases

Recent advances in neural image compression (NIC) have resulted in models which are starting to outperform traditional codecs. While this has led to growing excitement about using these methods in real-world applications, the successful adoption of any machine learning system (including NIC) in the wild requires it to generalize (and be robust) to unseen distribution shifts at deployment time. Unfortunately, current research lacks comprehensive datasets and informative tools to evaluate and understand compression performance in real-world settings. To bridge this crucial gap, first, this paper presents a comprehensive benchmark suite to evaluate the out-of-distribution (OOD) performance of image compression methods. Specifically, we design CLIC-C and Kodak-C by introducing 15 common corruptions to popular CLIC and Kodak benchmarks. Next, we propose spectrally inspired introspection tools to gain a deeper understanding of errors introduced by image compression methods as well as their OOD performance. To this end, we carry out a detailed performance comparison of the classical codec with various variants of NIC (e.g., original, variable rate, pruned), revealing intriguing findings that challenge our current understanding of the strengths and limitations of NIC. Finally, we corroborate our empirical findings with theoretical analysis, providing an in-depth view of the OOD performance of NIC. Our benchmarks, spectral introspection tools, and findings provide a crucial bridge to the real-world adoption of NIC. We hope that our work will propel future efforts in designing more robust and generalizable NIC methods.

neural networks, Variational Autoencoder, robustne↗