Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “continuous reconstruction”

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 163 records · Page 9
All sourcesNASA NTRSDOE OSTI
NASA NTRS1986

High resolution observations of the L1551 bipolar outflow

The nearby dark cloud Lynds 1551 contains one of the closest examples of a well-collimated bipolar molecular outflow. This source has the largest angular size of any known outflow and was the first bipolar outflow to be detected. The outflow originates from a low-luminosity young stellar object, IRS-5. Optical and radio continuum observations show the presence of a highly collimated, ionized stellar wind orginating from close to IRS-5 and aligned with the molecular outflow. However, we have little information on the actual mechanism that generates the stellar wind and collimates it into opposed jets. The Very Large Array (VLA) observations indicate that the winds originate within 10(15) cm of IRS-5, unfortunately at a size scale difficult to resolve. For these reasons, observations of the structure and dynamics of the hypersonic molecular gas may provide valuable information on the origin and evolution of these outflows. In addition, the study of the impact of the outflowing gas on the surrounding molecular material is essential to understand the consequence these outflows have on the evolution and star formation history of the entire cloud. Moriarty-Schieven et al. (1986) obtained a oversampled map of the CO emission of a portion of both the blueshifted and redshifted outflows in LI551 using Five College Radio Astronomy Observatory 14 m telescope. The oversampled maps have been reconstructed to an effective angular resolution of 20 arcsec using a maximum entropy algorithm. A continuation of the study of Moriarty-Schieven et al. is presented. The entire L1551 outflow has now been mapped at 12 arcsec sampling requiring roughly 4000 spectra. This data has been constructed to 20 arcsec resolution to provide the first high resolution picture of the entire L1551 outflow. This new data has shown that the blueshifted lobe is more extended than previously thought and has expanded downstream sufficiently to break out of the dense molecular cloud, but the redshifted outflow is still confined within the molecular cloud. Details of the structure and kinematics of the high velocity gas are used to test the various models of the origin and evolution of outflows.

Snell, R.↗
NASA NTRS2023

How do tradeoffs in satellite spatial and temporal resolution impact snow water equivalent reconstruction?

Given the tradeoffs between spatial and temporal resolution, questions about resolution optimality are fundamental to the study of global snow. Answers to these questions will inform future scientific priorities and mission specifications. Heterogeneity of mountain snowpacks drives a need for daily snow cover mapping at the slope scale (≤30 m) that is unmet for a variety of scientific users, ranging from hydrologists to the military to wildlife biologists. But finer spatial resolution usually requires coarser temporal or spectral resolution. Thus, no single sensor can meet all these needs. Recently, constellations of satellites and fusion techniques have made noteworthy progress. The efficacy of two such recent advances is examined: (1) a fused MODIS–Landsat product with daily 30 m spatial resolution and (2) a harmonized Landsat 8 and Sentinel 2A and B (HLS) product with 3–4 d temporal and 30 m spatial resolution. State-of-the-art spectral unmixing techniques are applied to surface reflectance products from 1 and 2 to create snow cover and albedo maps. Then an energy balance model was run to reconstruct snow water equivalent (SWE). For validation, lidar-based Airborne Snow Observatory SWE estimates were used. Results show that reconstructed SWE forced with 30 m resolution snow cover has lower bias, a measure of basin-wide accuracy, than the baseline case using MODIS (463 m cell size) but greater mean absolute error, a measure of per-pixel accuracy. However, the differences in errors may be within uncertainties from scaling artifacts, e.g., basin boundary delineation. Other explanations are (1) the importance of daily acquisitions and (2) the limitations of downscaled forcings for reconstruction. Conclusions are as follows: (1) spectrally unmixed snow cover and snow albedo from MODIS continue to provide accurate forcings for snow models and (2) finer spatial and temporal resolution through sensor design, fusion techniques, and satellite constellations are the future for Earth observations, but existing moderate-resolution sensors still offer value.

Edward H. Bair↗
DOE OSTI2016

Neural speech recognition: continuous phoneme decoding using spatiotemporal representations of human cortical activity

The superior temporal gyrus (STG) and neighboring brain regions play a key role in human language processing. Previous studies have attempted to reconstruct speech information from brain activity in the STG, but few of them incorporate the probabilistic framework and engineering methodology used in modern speech recognition systems. Here, in this work, we describe the initial efforts toward the design of a neural speech recognition (NSR) system that performs continuous phoneme recognition on English stimuli with arbitrary vocabulary sizes using the high gamma band power of local field potentials in the STG and neighboring cortical areas obtained via electrocorticography. The system implements a Viterbi decoder that incorporates phoneme likelihood estimates from a linear discriminant analysis model and transition probabilities from an n-gram phonemic language model. Grid searches were used in an attempt to determine optimal parameterizations of the feature vectors and Viterbi decoder. The performance of the system was significantly improved by using spatiotemporal representations of the neural activity (as opposed to purely spatial representations) and by including language modeling and Viterbi decoding in the NSR system. These results emphasize the importance of modeling the temporal dynamics of neural responses when analyzing their variations with respect to varying stimuli and demonstrate that speech recognition techniques can be successfully leveraged when decoding speech from neural signals. Guided by the results detailed in this work, further development of the NSR system could have applications in the fields of automatic speech recognition and neural prosthetics.

59 BASIC BIOLOGICAL SCIENCES↗
NASA NTRS2018

Multi-frequency Tomography Radar Observations of Snow Stratigraphy at Fraser during SnowEx

SnowEx is a multi-year airborne snow campaign led by NASA. The purpose of SnowEx is to figure out how much water is stored in Earth’s terrestrial snow-covered regions. As part of the 2017 NASA SnowEx campaign, we deployed a portable triple-frequency (9.6GHz, 13.5GHz and 17.2GHz) and fully polarimetric frequency-modulated continuous-wave (FMCW) radar at Fraser, Colorado. The radar was installed on a 60cmx60cm frame to enable a full reconstruction of the three-dimensional variability per each radar channel. The tomography technique uses the radar echo from the multiple viewing positions and provides a unique access to the vertical structure of the snow layer. With current setup, the range resolution is 30cm. In this paper, we will review the radar design and signal-processing algorithm – time domain back projection. The generated vertical images show the snow stratigraphy, which is consistent with ground snow pit measurement. The continuous operation demonstrates diurnal thawing and refreezing process. The snow density is retrieved by comparing to the snow free image.

Esteban-Fernandez, Daniel↗
DOE OSTI2015

A general reconstruction of the recent expansion history of the universe

Distance measurements are currently the most powerful tool to study the expansion history of the universe without specifying its matter content nor any theory of gravitation. Assuming only an isotropic, homogeneous and flat universe, in this work we introduce a model-independent method to reconstruct directly the deceleration function via a piecewise function. Including a penalty factor, we are able to vary continuously the complexity of the deceleration function from a linear case to an arbitrary (n+1)-knots spline interpolation. We carry out a Monte Carlo (MC) analysis to determine the best penalty factor, evaluating the bias-variance trade-off, given the uncertainties of the SDSS-II and SNLS supernova combined sample (JLA), compilations of baryon acoustic oscillation (BAO) and H(z) data. The bias-variance analysis is done for three fiducial models with different features in the deceleration curve. We perform the MC analysis generating mock catalogs and computing their best-fit. For each fiducial model, we test different reconstructions using, in each case, more than 10{sup 4} catalogs in a total of about 5× 10{sup 5}. This investigation proved to be essential in determining the best reconstruction to study these data. We show that, evaluating a single fiducial model, the conclusions about the bias-variance ratio are misleading. We determine the reconstruction method in which the bias represents at most 10% of the total uncertainty. In all statistical analyses, we fit the coefficients of the deceleration function along with four nuisance parameters of the supernova astrophysical model. For the full sample, we also fit H{sub 0} and the sound horizon r{sub s}(z{sub d}) at the drag redshift. The bias-variance trade-off analysis shows that, apart from the deceleration function, all other estimators are unbiased. Finally, we apply the Ensemble Sampler Markov Chain Monte Carlo (ESMCMC) method to explore the posterior of the deceleration function up to redshift 1.3 (using only JLA) and 2.3 (JLA+BAO+H(z)). We obtain that the standard cosmological model agrees within 3σ level with the reconstructed results in the whole studied redshift intervals. Since our method is calibrated to minimize the bias, the error bars of the reconstructed functions are a good approximation for the total uncertainty.

79 ASTRONOMY AND ASTROPHYSICS↗
NASA NTRS2021

The role of analysis error in the convergence of reanalysis production streams in MERRA-2

Due to production time constraints, most reanalyses are produced in multiple parallel streams instead of a single continuous one. These streams cover separate segments of the reanalysis time period with short overlaps to allow reconstruction of the official record. A fundamental assumption justifying this approach is that the streams will be assimilating the same observations during the periods where they overlap, and so will eventually converge to a similar atmospheric state, making discontinuities at stream junctions negligible. This assumption is revisited in this work by examining the impact of analysis error on the differences between MERRA-2 overlapping streams in three historical periods. Comparison results are shown in terms of standard deviations of stream differences as well as the spectral decomposition of the variance of their differences. Residual differences were found at the end of each year of overlap, with larger values observed in the earlier segments of the presatellite era. By drawing parallels with analysis error statistics estimated from the GMAO OSSE system, these differences are shown to reflect the varying constraint of data with the varying observing network, and to further carry the imprint of errors that the data assimilation process is not able to mitigate. As such, they are unlikely to be reduced by longer spinup periods. The ability of data assimilation to ensure continuity in the parallel streams is put into question when the observing system coverage is inadequate or simply when the data assimilation system as a whole is suboptimal.

Amal El Akkraoui↗
DOE OSTI2026

Extensive analysis of reconstruction algorithms for DESI 2024 baryon acoustic oscillations

Reconstruction of the baryon acoustic oscillation (BAO) signal has been a standard procedure in BAO analyses over the past decade and has helped to improve the BAO parameter precision by a factor of ∼2 on average. The Dark Energy Spectroscopic Instrument (DESI) BAO analysis for the first year (DR1) data uses the “standard” reconstruction framework, in which the displacement field is estimated from the observed density field by solving the linearized continuity equation in redshift space, and galaxy and random positions are shifted in order to partially remove non-linearities. There are several approaches to solving for the displacement field in real survey data, including the multigrid (MG), iterative Fast Fourier Transform (iFFT), and iterative Fast Fourier Transform particle (iFFTP) algorithms. In this work, we analyze these algorithms and compare them with various metrics including two-point statistics and the displacement itself using realistic DESI mocks. We focus on three representative DESI samples, the emission line galaxies (ELG), quasars (QSO), and the bright galaxy sample (BGS), which cover the extreme redshifts and number densities, and potential wide-angle effects. We conclude that the MG and iFFT algorithms agree within 0.4% in post-reconstruction power spectrum on BAO scales with the RecSym convention, which does not remove large-scale redshift space distortions (RSDs), in all three tracers. The RecSym convention appears to be less sensitive to displacement errors than the RecIso convention, which attempts to remove large-scale RSDs. However, iFFTP deviates from the first two; thus, we recommend against using iFFTP without further development. In addition, we provide the optimal settings for reconstruction for five years of DESI observation. The analyses presented in this work pave the way for DESI DR1 analysis as well as future BAO analyses.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2026

3D reconstruction and neural rendering for adversarial machine learning

While evasion attacks on computer vision systems have been widely studied, creating attacks that remain effective under significant changes in viewpoint continues to be challenging. Traditional approaches often rely on affine transformations of images, but these approaches degrade at larger perspective shifts and often produce unrealistic or ineffective perturbations. Recent methods use differentiable renderers to improve viewpoint robustness, but they typically depend on manually constructed 3D models. We introduce a semi-automated pipeline that generates physically printable and perspective-invariant adversarial patches using only a small set of 2D images. Our method integrates 3D reconstruction, neural rendering, adversarial patch optimization, and an object detection victim model into a unified workflow. We use 2D Gaussian Splatting for high fidelity mesh reconstruction and FlexPara for surface parameterization that produces texture maps suitable for patch editing. Together, these components form a fully differentiable pipeline in PyTorch3D that links texture modification to model outputs, enabling efficient optimization of patches that remain effective across many viewpoints. The complete process, from image capture to patch printing and physical evaluation, can be completed within a few hours. We demonstrate the effectiveness of the resulting patches through attacks on the YOLOv8 object detection model and discuss remaining challenges and opportunities for improving robustness and scalability.

Singhvi, Vivaan [ORNL] (ORCID:0009000586288221)↗
DOE OSTI2018

Dual-Energy CT Pulmonary Angiography: Quantification of Disease Burden and Impact on Management

Purpose of Review: Computed tomography pulmonary angiography (CTPA) has become the imaging modality of choice for patients with suspected pulmonary embolism (PE). Post-processing techniques currently available for dual-energy CT pulmonary angiography (DE-CTPA) enhance image quality and provide additional value in the diagnosis of PE. The objective of this article is to summarize these recent developments and discuss the appropriate use of DE-CTPA post-processing applications. Recent Findings: DE-CTPA post-processing applications enable reconstruction of virtual monoenergetic images (VMI) and color-coded iodine-perfusion maps to increase contrast conditions and visualize lung perfusion defects in case of embolic occlusion of pulmonary arteries. Both techniques revealed a superior diagnostic performance for the detection of pulmonary emboli and assessment of the pulmonary perfusion compared to the standard image reconstructions. Summary: DE-CTPA is a well-established method for excluding or diagnosing PE. Continued developments in DE-CTPA post-processing techniques improve patient management and allow for a quantification of disease burden.

62 RADIOLOGY AND NUCLEAR MEDICINE↗
DOE OSTI2019

Mooring and Seafloor Pressure End Point Measurements at the Southern Entrance of the Solomon Sea: Subseasonal to Interannual Flow Variability

Variability of the flow across the Solomon Sea's southern entrance was examined using end point subsurface moorings and seafloor pressure sensors, reconstructed velocity profiles based on satellite-derived surface velocity and bottom pressure-derived subsurface velocity, and 1993–2017 proxy volume transport based on satellite altimetry. The reconstructed velocity correctly represents the fluctuating surface flow and subsurface core providing a high-frequency continuous observing system for this sea. The mean equatorward volume transport over 0- to 500-m depth layer is 15.2 Sv (1 Sv ≡ 106 m3/s) during July 2012 to May 2017. The measurements resolve the full spectrum of the volume transport including energetic subseasonal variability that fluctuates by as much as 25 Sv over one week. At low-frequency timescales, the study finds that linear Rossby waves forced by Ekman pumping in the interior of the Pacific influence not only seasonal fluctuations as found by previous studies but also interannual variability. As found previously, the El Niño–Southern Oscillation highly influences interannual volume transport. During the 2015/2016 El Niño, observations show the seasonal cycle to be suppressed from the second half of 2014, prior to the mature phase of the El Niño, to September 2016 along with an increase in across-transect transport. At subseasonal timescales, local Ekman pumping and remote wind stress curl are responsible for a third of the subseasonal variance. Finally, the study highlights the importance of high-frequency observations at the southern entrance of the Solomon Sea and the ability of a linear Rossby model to represent the low-frequency variability of the transport.

54 ENVIRONMENTAL SCIENCES↗
NASA NTRS1988

Wiener restoration of sampled image data - End-to-end analysis

The Wiener filter is formulated as a function of the basic image-gathering and image-reconstruction constraints, thereby providing a method for minimizing the mean-squared error between the (continuous-input) radiance field and its restored (continuous-output) representation. This formulation of the Wiener filter is further extended to the Wiener-characteristic filter, which provides a method for explicitly specifying the desired representation. Two specific examples of Wiener filters are presented.

Fales, Carl L.↗
DOE OSTI2016

SU-F-J-175: Evaluation of Metal Artifact Reduction Algorithms in Computed Tomography and Their Application to Radiation Therapy Treatment Planning

Purpose: High-Z (metal) implants in CT scans cause significant streak-like artifacts in the reconstructed dataset. This results in both inaccurate CT Hounsfield units for the tissue as well as obscuration of the target and organs at risk (OARs) for radiation therapy planning. Herein we analyze two metal artifact reduction algorithms: GE’s Smart MAR and a Metal Deletion Technique (MDT) for geometric and Hounsfield Unit (HU) accuracy. Methods: A CT-to-electron density phantom, with multiple inserts of various densities and a custom Cerrobend insert (Zeff=76.8), is utilized in this continuing study. The phantom is scanned without metal (baseline) and again with the metal insert. Using one set of projection data, reconstructed CT volumes are created with filtered-back-projection (FBP) and the MAR and the MDT algorithms. Regions-of-Interest (ROIs) are evaluated for each insert for HU accuracy; the metal insert’s Full-Width-Half-Maximum (FWHM) is used to evaluate the geometric accuracy. Streak severity is quantified with an HU error metric over the phantom volume. Results: The original FBP reconstruction has a Root-Mean-Square-Error (RMSE) of 57.55 HU (STD=29.19, range=−145.8 to +79.2) compared to baseline. The MAR reconstruction has a RMSE of 20.98 HU (STD=13.92, range=−18.3 to +61.7). The MDT reconstruction has a RMSE of 10.05 HU (STD=10.5, range=−14.8 to +18.6). FWHM for baseline=162.05; FBP=161.84 (−0.13%); MAR=162.36 (+0.19%); MDT=162.99 (+0.58%). Streak severity metric for FBP=19.73 (22.659% bad pixels); MAR=8.743 (9.538% bad); MDT=4.899 (5.303% bad). Conclusion: Image quality, in terms of HU accuracy, in the presence of high-Z metal objects in CT scans is improved by metal artifact reduction reconstruction algorithms. The MDT algorithm had the highest HU value accuracy (RMSE=10.05 HU) and best streak severity metric, but scored the worst in terms of geometric accuracy. Qualitatively, the MAR and MDT algorithms increased detectability of inserts, although there is a loss of in-plane resolution near the metallic insert.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2015

Characterization of a detector chain using a FPGA-based time-to-digital converter to reconstruct the three-dimensional coordinates of single particles at high flux

We report on the development of a novel FPGA-based time-to-digital converter and its implementation in a detection chain that records the coordinates of single particles along three dimensions. The detector is composed of micro-channel plates mounted on top of a cross delay line and connected to fast electronics. We demonstrate continuous recording of the timing signals from the cross delay line at rates up to 4.1 × 10{sup 6} s{sup −1} and three-dimensional reconstruction of the coordinates up to 3.2 × 10{sup 6} particles per second. From the imaging of a calibrated structure we measure the in-plane resolution of the detector to be 140(20) μm at a flux of 3 × 10{sup 5} particles per second. In addition, we analyze a method to estimate the resolution without placing any structure under vacuum, a significant practical improvement. While we use UV photons here, the results of this work apply to the detection of other kinds of particles.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2025

Investigating the Global Biogeophysical Impact of Area and Mass Based Wood Harvest in a Vegetation Demography Model

Wood harvesting alters land surface properties and energy redistribution, but there is a lack of studies estimating these changes on a global scale. We coupled a vegetation demographic model, the Functionally Assembled Terrestrial Ecosystem Simulator, with the E3SM land model to perform offline model simulation to investigate the land biogeophysical responses, including canopy coverage, leaf area index, albedo, surface roughness length, and energy fluxes, to historical wood harvest on the global scale. In this study, we found 50% less harvested carbon (C) when choosing the area-based harvest rate as driving data that has not been spatially harmonized, compared to reharmonized mass-based harvesting. By considering the uncertainty from reconstruction of historical wood harvest time series and the choice of wood harvest approach in the model, continuous wood harvest (1850–2015) results in 5%–10% of canopy coverage loss, contributing 0.5%–1% increase of albedo over disturbed land, which is much stronger than a non-demographic land surface model. Changes in energy flux from the wood harvest are negligible (<1%), but the responses of land surface properties vary (up to 30%) due to differences in model structure between the single canopy, sun-shade leaf model and vegetation demographic model.

Shu, Shijie [Lawrence Berkeley National Laboratory↗
DOE OSTI2022

Observation of radon mitigation in MicroBooNE by a liquid argon filtration system

We report the MicroBooNE liquid argon time projection chamber (LArTPC) maintains a high level of liquid argon purity through the use of a filtration system that removes electronegative contaminants in continuously-circulated liquid, recondensed boil off, and externally supplied argon gas. We use the MicroBooNE LArTPC to reconstruct MeV-scale radiological decays. Using this technique we measure the liquid argon filtration system's efficacy at removing radon. This is studied by placing a 500 kBq 222 Rn source upstream of the filters and searching for a time-dependent increase in the number of radiological decays in the LArTPC. In the context of two models for radon mitigation via a liquid argon filtration system, a slowing mechanism and a trapping mechanism, MicroBooNE data supports a radon reduction factor of greater than 97% or 99.999%, respectively. Furthermore, a radiological survey of the filters found that the copper-based filter material was the primary medium that removed the 222 Rn. This is the first observation of radon mitigation in liquid argon with a large-scale copper-based filter and could offer a radon mitigation solution for future large LArTPCs.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2018

Sparse Coding for N-Gram Feature Extraction and Training for File Fragment Classification

File fragment classification is an important step in the task of file carving in digital forensics. In file carving, files must be reconstructed based on their content as a result of their fragmented storage on disk or in memory. Existing methods for classification of file fragments typically use hand-engineered features such as byte histograms or entropy measures. In this paper, we propose an approach using sparse coding that enables automated feature extraction. Sparse coding, or sparse dictionary learning, is an unsupervised learning algorithm, and is capable of extracting features based simply on how well those features can be used to reconstruct the original data. With respect to file fragments, we learn sparse dictionaries for n-grams, continuous sequences of bytes, of different sizes. These dictionaries may then be used to estimate n-gram frequencies for a given file fragment, but for significantly larger n-gram sizes than are typically found in existing methods which suffer from combinatorial explosion. To demonstrate the capability of our sparse coding approach, we used the resulting features to train standard classifiers such as support vector machines (SVMs) over multiple file types. Experimentally, we achieved significantly better classification results with respect to existing methods, especially when the features were used in supplement to existing hand-engineered features.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2016

AN ASSESSMENT OF SUNSPOT NUMBER DATA COMPOSITES OVER 1845–2014

New sunspot data composites, some of which are radically different in the character of their long-term variation, are evaluated over the interval 1845–2014. The method commonly used to calibrate historic sunspot data, relative to modern-day data, is “daisy-chaining,” whereby calibration is passed from one data subset to the neighboring one, usually using regressions of the data subsets for the intervals of their overlap. Recent studies have illustrated serious pitfalls in these regressions, and the resulting errors can be compounded by their repeated use as the data sequence is extended back in time. Hence, the recent composite data series by Usoskin et al., R {sub UEA}, is a very important advance because it avoids regressions, daisy-chaining, and other common, but invalid, assumptions: this is achieved by comparing the statistics of “active-day” fractions to those for a single reference data set. We study six sunspot data series, including R {sub UEA} and the new “backbone” data series ( R {sub BB}, recently generated by Svalgaard and Schatten by employing both regression and daisy-chaining). We show that all six can be used with a continuity model to reproduce the main features of the open solar flux variation for 1845–2014, as reconstructed from geomagnetic activity data. However, some differences can be identified that are consistent with tests using a basket of other proxies for solar magnetic fields. Using data from a variety of sunspot observers, we illustrate problems with the method employed in generating R {sub BB} that cause it to increasingly overestimate sunspot numbers going back in time, and we recommend using R {sub UEA} because it employs more robust procedures that avoid such problems.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2025

Joint Modeling of Quasar Variability and Accretion Disk Reprocessing Using Latent Stochastic Differential Equations

Quasars are bright active galactic nuclei powered by the accretion of matter around supermassive black holes at the center of galaxies. Their stochastic brightness variability depends on the physical properties of the accretion disk and black hole. The upcoming Rubin Observatory Legacy Survey of Space and Time (LSST) is expected to observe tens of millions of quasars, so there is a need for efficient techniques like machine learning that can handle the large volume of data. Quasar variability is believed to be driven by an X-ray corona, which is reprocessed by the accretion disk and emitted as UV/optical variability. We are the first to introduce an auto-differentiable simulation of the accretion disk and reprocessing. We use the simulation as a direct component of our neural network to jointly model the driving variability and reprocessing, trained with supervised learning on simulated LSST-like 10 yr quasar light curves. We encode the light curves using a transformer encoder, and the driving variability is reconstructed using latent stochastic differential equations, a physically motivated generative deep learning method that can model continuous-time stochastic dynamics. By embedding the physical processes of the driving signal and reprocessing into our network, we achieve a model that is more robust and interpretable. We demonstrate that our model outperforms a Gaussian process regression baseline and can infer accretion disk parameters and time delays between wave bands, even for out-of-distribution driving signals. Our approach provides a powerful framework that can be adapted to solve other inverse problems in multivariate time series.

Fagin, Joshua [City Univ. of New York (CUNY), NY (↗