arXiv subjects

Xudong Sun

Publications and source records attributed to Xudong Sun.

At least 55 records · Page 3Linked to original sources

arXiv2021

Torus-Stable Zone Above Starspots

Whilst intense solar flares are almost always accompanied by a coronal mass ejection (CME), reports on stellar CMEs are rare, despite the frequent detection of stellar 'super flares'. The torus instability of magnetic flux ropes is believed to be one of the main driving mechanisms of solar CMEs. Suppression of the torus instability, due to a confining background coronal magnetic field that decreases sufficiently slowly with height, may contribute to the lack of stellar CME detection. Here we use the solar magnetic field as a template to estimate the vertical extent of this 'torus-stable zone' (TSZ) above a stellar active region. For an idealised potential field model comprising the fields of a local bipole (mimicking a pair of starspots) and a global dipole, we show that the upper bound of the TSZ increases with the bipole size, the dipole strength, and the source surface radius where the coronal field becomes radial. The boundaries of the TSZ depend on the interplay between the spots' and the dipole's magnetic fields, which provide the local- and global-scale confinement, respectively. They range from about half the bipole size to a significant fraction of the stellar radius. For smaller spots and an intermediate dipole field, a secondary TSZ arises at a higher altitude, which may increase the likelihood of 'failed eruptions'. Our results suggest that the low apparent CME occurrence rate on cool stars is, at least partially, due to the presence of extended TSZs.

astro-ph.SR↗

arXiv2021

Rapid Evolution of Bald Patches in a Major Solar Eruption

Bald patch (BP) is a magnetic topological feature where U-shaped field lines turn tangent to the photosphere. Field lines threading the BP trace a separatrix surface where reconnection preferentially occurs. Here we study the evolution of multiple, strong-field BPs in active region 12673 during the most intense, X9.3 flare of solar cycle 24. The central BP, located between the initial flare ribbons, largely "disintegrated" within 35 minutes. The more remote, southern BP survived. The disintegration manifested as a 9-degree rotation of the median shear angle; the perpendicular component of the horizontal field (with respect to the polarity inversion line) changed sign. The parallel component exhibited a step-wise, permanent increase of 1 kG, consistent with previous observations of the flare-related "magnetic imprint". The observations suggest that magnetic reconnection during a major eruption may involve entire BP separatrices, leading to a change of magnetic topology from BPs to sheared arcades.

astro-ph.SR↗

arXiv2021

SMARPs and SHARPs: Two Solar Cycles of Active Region Data

We present a new data product, called Space-Weather MDI Active Region Patches (SMARPs), derived from maps of the solar surface magnetic field taken by the Michelson Doppler Imager (MDI) aboard the Solar and Heliospheric Observatory (SoHO). Together with the Space-Weather HMI Active Region Patches (SHARPs), derived from similar maps taken by the Helioseismic and Magnetic Imager (HMI) aboard the Solar Dynamics Observatory, these data provide a continuous and seamless set of maps and keywords that describe every active region observed over the last two solar cycles, from 1996 to the present day. In this paper, we describe the SMARP data and compare it to the SHARP data.

astro-ph.SR↗

arXiv2021

Are the Magnetic Fields Radial in the Solar Polar Region?

We investigate the orientation of the photospheric magnetic fields in the solar polar region using observations from the Helioseismic and Magnetic Imager (HMI). Inside small patches of significant polarization, the inferred magnetic field vectors at $1''$ scale appear to systematically deviate from the radial direction. Most tilt towards the pole; all are more inclined toward the plane of sky compared to the radial vector. These results, however, depend on the "filling factor" $f$ that characterizes the unresolved magnetic structures. The default, uninformative $f\equiv1$ for HMI will incur larger inclination and less radial fields than $f<1$. The observed trend may be a systematic bias inherent to the limited resolution.

astro-ph.SR↗

arXiv2021

Hierarchical Variational Auto-Encoding for Unsupervised Domain Generalization

We address the task of domain generalization, where the goal is to train a predictive model such that it is able to generalize to a new, previously unseen domain. We choose a hierarchical generative approach within the framework of variational autoencoders and propose a domain-unsupervised algorithm that is able to generalize to new domains without domain supervision. We show that our method is able to learn representations that disentangle domain-specific information from class-label specific information even in complex settings where domain structure is not observed during training. Our interpretable method outperforms previously proposed generative algorithms for domain generalization as well as other non-generative state-of-the-art approaches in several hierarchical domain settings including sequential overlapped near continuous domain shift. It also achieves competitive performance on the standard domain generalization benchmark dataset PACS compared to state-of-the-art approaches which rely on observing domain-specific information during training, as well as another domain unsupervised method. Additionally, we proposed model selection purely based on Evidence Lower Bound (ELBO) and also proposed weak domain supervision where implicit domain information can be added into the algorithm.

cs.LG↗

arXiv2021

Grid-GCN for Fast and Scalable Point Cloud Learning

Due to the sparsity and irregularity of the point cloud data, methods that directly consume points have become popular. Among all point-based models, graph convolutional networks (GCN) lead to notable performance by fully preserving the data granularity and exploiting point interrelation. However, point-based networks spend a significant amount of time on data structuring (e.g., Farthest Point Sampling (FPS) and neighbor points querying), which limit the speed and scalability. In this paper, we present a method, named Grid-GCN, for fast and scalable point cloud learning. Grid-GCN uses a novel data structuring strategy, Coverage-Aware Grid Query (CAGQ). By leveraging the efficiency of grid space, CAGQ improves spatial coverage while reducing the theoretical time complexity. Compared with popular sampling methods such as Farthest Point Sampling (FPS) and Ball Query, CAGQ achieves up to 50X speed-up. With a Grid Context Aggregation (GCA) module, Grid-GCN achieves state-of-the-art performance on major point cloud classification and segmentation benchmarks with significantly faster runtime than previous studies. Remarkably, Grid-GCN achieves the inference speed of 50fps on ScanNet using 81920 points per scene as input.

cs.CV↗

arXiv2021

Proxy-Based Prediction of Solar Extreme Ultraviolet Emission using Deep Learning

High-energy radiation from the Sun governs the behavior of Earth's upper atmosphere and such radiation from any planet-hosting star can drive the long-term evolution of a planetary atmosphere. However, much of this radiation is unobservable because of absorption by Earth's atmosphere and the interstellar medium. This motivates the identification of a proxy that can be readily observed from the ground. Here, we evaluate absorption in the near-infrared 1083 nm triplet line of neutral orthohelium as a proxy for extreme ultraviolet (EUV) emission in the 30.4 nm line of He II and 17.1 nm line of Fe IX from the Sun. We apply deep learning to model the non-linear relationships, training and validating the model on historical, contemporaneous images of the solar disk acquired in the triplet He I line by the ground-based SOLIS observatory and in the EUV by the NASA Solar Dynamics Observatory. The model is a fully-convolutional neural network (FCN) that incorporates spatial information and accounts for the projection of the spherical Sun to 2-d images. Using normalized target values, results indicate a median pixel-wise relative error of 20% and a mean disk-integrated flux error of 7% on a held-out test set. Qualitatively, the model learns the complex spatial correlations between He I absorption and EUV emission has a predictive ability superior to that of a pixel-by-pixel model; it can also distinguish active regions from high-absorption filaments that do not result in EUV emission.

astro-ph.SR↗

arXiv2021

Emerging Dimming as Coronal Heating Episodes

Emerging dimming occurs in isolated solar active regions (ARs) during the early stages of magnetic flux emergence. Observed by the Atmospheric Imaging Assembly, it features a rapid decrease in extreme-ultraviolet (EUV) emission in the 171 Å channel images, and a simultaneous increase in the 211 Å images. Here, we analyze the coronal thermodynamic and magnetic properties to probe its physical origin. We calculate the time-dependent differential emission measures for a sample of 18 events between 2010 and 2012. The emission measure (EM) decrease in the temperature range $5.7 \le \log_{10}T \le 5.9$ is well correlated with the EM increase in $6.2 \le \log_{10}T \le 6.4$ over eight orders of magnitude. This suggests that the coronal plasma is being heated from the quiet-Sun, sub-MK temperature to 1-2 MK, more typical for ARs. Potential field extrapolation indicates significant change in the local magnetic connectivity: the dimming region is now linked to the newly emerged flux via longer loops. We conclude that emerging dimming is likely caused by coronal heating episodes, powered by reconnection between the emerging and the ambient magnetic fields.

astro-ph.SR↗

arXiv2021

Benchmarking time series classification -- Functional data vs machine learning approaches

Time series classification problems have drawn increasing attention in the machine learning and statistical community. Closely related is the field of functional data analysis (FDA): it refers to the range of problems that deal with the analysis of data that is continuously indexed over some domain. While often employing different methods, both fields strive to answer similar questions, a common example being classification or regression problems with functional covariates. We study methods from functional data analysis, such as functional generalized additive models, as well as functionality to concatenate (functional-) feature extraction or basis representations with traditional machine learning algorithms like support vector machines or classification trees. In order to assess the methods and implementations, we run a benchmark on a wide variety of representative (time series) data sets, with in-depth analysis of empirical results, and strive to provide a reference ranking for which method(s) to use for non-expert practitioners. Additionally, we provide a software framework in R for functional data analysis for supervised learning, including machine learning and more linear approaches from statistics. This allows convenient access, and in connection with the machine-learning toolbox mlr, those methods can now also be tuned and benchmarked.

stat.ML↗

arXiv2020

More Industry-friendly: Federated Learning with High Efficient Design

Although many achievements have been made since Google threw out the paradigm of federated learning (FL), there still exists much room for researchers to optimize its efficiency. In this paper, we propose a high efficient FL method equipped with the double head design aiming for personalization optimization over non-IID dataset, and the gradual model sharing design for communication saving. Experimental results show that, our method has more stable accuracy performance and better communication efficient across various data distributions than other state of art methods (SOTAs), makes it more industry-friendly.

cs.LG↗

arXiv2020

Non-Neutralized Electric Current of Active Regions Explained as a Projection Effect

Active regions (ARs) often possess an observed net electric current in a single magnetic polarity. We show that such "non-neutralized" currents can arise from a geometric projection effect when a twisted flux tube obliquely intersects the photosphere. To this end, we emulate surface maps of an emerging AR by sampling horizontal slices of a semi-torus flux tube at various heights. Although the tube has no net toroidal current, its poloidal current, when projected along the vertical direction, amounts to a significant non-neutralized component on the surface. If the tube emerges only partially as in realistic settings, the non-neutralized current will 1) develop as double ribbons near the sheared polarity inversion line, (2) positively correlate with the twist, and 3) reach its maximum before the magnetic flux. The projection effect may be important to the photospheric current distribution, in particular during the early stages of flux emergence.

astro-ph.SR↗

arXiv2020

The Coronal Global Evolutionary Model: Using HMI Vector Magnetogram and Doppler Data to Determine Coronal Magnetic Field Evolution

The Coronal Global Evolutionary Model (CGEM) provides data-driven simulations of the magnetic field in the solar corona to better understand the build-up of magnetic energy that leads to eruptive events. The CGEM project has developed six capabilities. CGEM modules (1) prepare time series of full-disk vector magnetic field observations to (2) derive the changing electric field in the solar photosphere over active-region scales. This local electric field is (3) incorporated into a surface flux transport model that reconstructs a global electric field that evolves magnetic flux in a consistent way. These electric fields drive a (4) 3D spherical magneto-frictional (SMF) model, either at high-resolution over a restricted range of solid angle or at lower resolution over a global domain, to determine the magnetic field and current density in the low corona. An SMF-generated initial field above an active region and the evolving electric field at the photosphere are used to drive (5) detailed magneto-hydrodynamic (MHD) simulations of active regions in the low corona. SMF or MHD solutions are then used to compute emissivity proxies that can be compared with coronal observations. Finally, a lower-resolution SMF magnetic field is used to initialize (6) a global MHD model that is driven by an SMF electric-field time series to simulate the outer corona and heliosphere, ultimately connecting Sun to Earth. As a demonstration, this report features results of CGEM applied to observations of the evolution of NOAA Active Region 11158 in February 2011.

astro-ph.SR↗

arXiv2020

Critical Science Plan for the Daniel K. Inouye Solar Telescope (DKIST)

The Daniel K. Inouye Solar Telescope (DKIST) will revolutionize our ability to measure, understand and model the basic physical processes that control the structure and dynamics of the Sun and its atmosphere. The first-light DKIST images, released publicly on 29 January 2020, only hint at the extraordinary capabilities which will accompany full commissioning of the five facility instruments. With this Critical Science Plan (CSP) we attempt to anticipate some of what those capabilities will enable, providing a snapshot of some of the scientific pursuits that the Daniel K. Inouye Solar Telescope hopes to engage as start-of-operations nears. The work builds on the combined contributions of the DKIST Science Working Group (SWG) and CSP Community members, who generously shared their experiences, plans, knowledge and dreams. Discussion is primarily focused on those issues to which DKIST will uniquely contribute.

astro-ph.SR↗

arXiv2020

Maximum Entropy-Regularized Multi-Goal Reinforcement Learning

In Multi-Goal Reinforcement Learning, an agent learns to achieve multiple goals with a goal-conditioned policy. During learning, the agent first collects the trajectories into a replay buffer, and later these trajectories are selected randomly for replay. However, the achieved goals in the replay buffer are often biased towards the behavior policies. From a Bayesian perspective, when there is no prior knowledge about the target goal distribution, the agent should learn uniformly from diverse achieved goals. Therefore, we first propose a novel multi-goal RL objective based on weighted entropy. This objective encourages the agent to maximize the expected return, as well as to achieve more diverse goals. Secondly, we developed a maximum entropy-based prioritization framework to optimize the proposed objective. For evaluation of this framework, we combine it with Deep Deterministic Policy Gradient, both with or without Hindsight Experience Replay. On a set of multi-goal robotic tasks of OpenAI Gym, we compare our method with other baselines and show promising improvements in both performance and sample-efficiency.

cs.LG↗

arXiv2020

The PDFI_SS Electric Field Inversion Software

We describe the PDFI_SS software library, which is designed to find the electric field at the Sun's photosphere from a sequence of vector magnetogram and Doppler velocity measurements, and estimates of horizontal velocities obtained from local correlation tracking using the recently upgraded FLCT code. The library, a collection of Fortran subroutines, uses the "PDFI" technique described by Kazachenko et al. (2014), but modified for use in spherical, Plate-Carrée geometry on a staggered grid. The domain over which solutions are found is a subset of the global spherical surface, defined by user-specified limits of colatitude and longitude. Our staggered-grid approach, based on that of Yee (1966), is more conservative and self-consistent compared to the centered, Cartesian grid used by Kazachenko et al. (2014). The library can be used to compute an end-to-end solution for electric fields from data taken by the HMI instrument aboard NASA's SDO Mission. This capability has been incorporated into the HMI pipeline processing system operating at SDO's JSOC. The library is written in a general and modular way so that the calculations can be customized to modify or delete electric field contributions, or used with other data sets. Other applications include "nudging" numerical models of the solar atmosphere to facilitate assimilative simulations. The library includes an ability to compute "global" (whole-Sun) electric field solutions. The library also includes an ability to compute Potential Magnetic Field solutions in spherical coordinates. This distribution includes a number of test programs which allow the user to test the software.

astro-ph.SR↗

arXiv2020

Electric Current Neutralization in Solar Active Regions and Its Relation to Eruptive Activity

It is well established that magnetic free energy associated with electric currents powers solar flares and coronal mass ejections (CMEs) from solar active regions (ARs). However, the conditions that determine whether an AR will produce an eruption are not well understood. Previous work suggests that the degree to which the driving electric currents, or the sum of all currents within a single magnetic polarity, are neutralized may serve as a good proxy for assessing the ability of ARs to produce eruptions. Here, we investigate the relationship between current neutralization and flare/CME production using a sample of 15 flare-active and 15 flare-quiet ARs. All flare-quiet and 4 flare-active ARs are also CME-quiet. We additionally test the relation of current neutralization to the degree of shear along polarity inversion lines (PILs) in an AR. We find that flare-productive ARs are more likely to exhibit non-neutralized currents, specifically those that also produce a CME. We find that flare/CME-active ARs also exhibit higher degrees of PIL shear than flare/CME-quiet ARs. We additionally observe that currents become more neutralized during magnetic flux emergence in flare-quiet ARs. Our investigation suggests that current neutralization in ARs is indicative of their eruptive potential.

astro-ph.SR↗

arXiv2019

Tutorial and Survey on Probabilistic Graphical Model and Variational Inference in Deep Reinforcement Learning

Aiming at a comprehensive and concise tutorial survey, recap of variational inference and reinforcement learning with Probabilistic Graphical Models are given with detailed derivations. Reviews and comparisons on recent advances in deep reinforcement learning are made from various aspects. We offer detailed derivations to a taxonomy of Probabilistic Graphical Model and Variational Inference methods in deep reinforcement learning, which serves as a complementary material on top of the original contributions.

cs.LG↗

arXiv2019

Variational Resampling Based Assessment of Deep Neural Networks under Distribution Shift

A novel variational inference based resampling framework is proposed to evaluate the robustness and generalization capability of deep learning models with respect to distribution shift. We use Auto Encoding Variational Bayes to find a latent representation of the data, on which a Variational Gaussian Mixture Model is applied to deliberately create distribution shift by dividing the dataset into different clusters. Wasserstein distance is used to characterize the extent of distribution shift between the generated data splits. We compare several popular Convolutional Neural Network (CNN) architectures and Bayesian CNN models for image classification on the Fashion-MNIST dataset, to assess their robustness and generalization behavior under the deliberately created distribution shift, as well as under random Cross Validation. Our method of creating artificial domain splits of a single dataset can also be used to establish novel model selection criteria and assessment tools in machine learning, as well as benchmark methods for domain adaptation and domain generalization approaches.

cs.LG↗