arXiv subjects

Xinming Wu

Publications and source records attributed to Xinming Wu.

28 records · Page 2Linked to original sources

arXiv2023

Seismic Foundation Model (SFM): a new generation deep learning model in geophysics

While computer science has seen remarkable advancements in foundation models, which remain underexplored in geoscience. Addressing this gap, we introduce a workflow to develop geophysical foundation models, including data preparation, model pre-training, and adaption to downstream tasks. From 192 globally collected 3-D seismic volumes, we create a carefully curated dataset of 2,286,422 2-D seismic images. Fully using these unlabeled images, we employ the self-supervised learning to pre-train a Transformer-based Seismic Foundation Model (SFM) for producing all-purpose seismic features that work across various tasks and surveys. Through experiments on seismic facies classification, geobody identification, interpolation, denoising, and inversion, our pre-trained model demonstrates versatility, generalization, scalability, and superior performance over baseline models. Conclusively, we provide a foundation model and vast dataset to advance AI in geophysics, addressing challenges (poor generalization, lacking labels, and repetitive training for task-specified models) of applying AI in geophysics and paving the way for future innovations in geoscience.

physics.geo-ph↗

arXiv2023

SeisCLIP: A seismology foundation model pre-trained by multi-modal data for multi-purpose seismic feature extraction

Training specific deep learning models for particular tasks is common across various domains within seismology. However, this approach encounters two limitations: inadequate labeled data for certain tasks and limited generalization across regions. To address these challenges, we develop SeisCLIP, a seismology foundation model trained through contrastive learning from multi-modal data. It consists of a transformer encoder for extracting crucial features from time-frequency seismic spectrum and an MLP encoder for integrating the phase and source information of the same event. These encoders are jointly pre-trained on a vast dataset and the spectrum encoder is subsequently fine-tuned on smaller datasets for various downstream tasks. Notably, SeisCLIP's performance surpasses that of baseline methods in event classification, localization, and focal mechanism analysis tasks, employing distinct datasets from different regions. In conclusion, SeisCLIP holds significant potential as a foundational model in the field of seismology, paving the way for innovative directions in foundation-model-based seismology research.

physics.geo-ph↗

arXiv2023

Multi-task multi-station earthquake monitoring: An all-in-one seismic Phase picking, Location, and Association Network (PLAN)

Earthquake monitoring is vital for understanding the physics of earthquakes and assessing seismic hazards. A standard monitoring workflow includes the interrelated and interdependent tasks of phase picking, association, and location. Although deep learning methods have been successfully applied to earthquake monitoring, they mostly address the tasks separately and ignore the geographic relationships among stations. Here, we propose a graph neural network that operates directly on multi-station seismic data and achieves simultaneous phase picking, association, and location. Particularly, the inter-station and inter-task physical relationships are informed in the network architecture to promote accuracy, interpretability, and physical consistency among cross-station and cross-task predictions. When applied to data from the Ridgecrest region and Japan regions, this method showed superior performance over previous deep learning-based phase-picking and localization methods. Overall, our study provides for the first time a prototype self-consistent all-in-one system of simultaneous seismic phase picking, association, and location, which has the potential for next-generation autonomous earthquake monitoring.

physics.geo-ph↗

arXiv2023

A Deep-Learning-Based Neural Decoding Framework for Emotional Brain-Computer Interfaces

Reading emotions precisely from segments of neural activity is crucial for the development of emotional brain-computer interfaces. Among all neural decoding algorithms, deep learning (DL) holds the potential to become the most promising one, yet progress has been limited in recent years. One possible reason is that the efficacy of DL strongly relies on training samples, yet the neural data used for training are often from non-human primates and mixed with plenty of noise, which in turn mislead the training of DL models. Given it is difficult to accurately determine animals' emotions from humans' perspective, we assume the dominant noise in neural data representing different emotions is the labeling error. Here, we report the development and application of a neural decoding framework called Emo-Net that consists of a confidence learning (CL) component and a DL component. The framework is fully data-driven and is capable of decoding emotions from multiple datasets obtained from behaving monkeys. In addition to improving the decoding ability, Emo-Net significantly improves the performance of the base DL models, making emotion recognition in animal models possible. In summary, this framework may inspire novel understandings of the neural basis of emotion and drive the realization of close-loop emotional brain-computer interfaces.

cs.HC↗

arXiv2022

Deep learning based sferics recognition for AMT data processing in the dead band

In the audio magnetotellurics (AMT) sounding data processing, the absence of sferic signals in some time ranges typically results in a lack of energy in the AMT dead band, which may cause unreliable resistivity estimate. We propose a deep convolutional neural network (CNN) to automatically recognize sferic signals from redundantly recorded data in a long time range and use them to compensate for the resistivity estimation. We train the CNN by using field time series data with different signal to noise rations that were acquired from different regions in mainland China. To solve the potential overfitting problem due to the limited number of sferic labels, we propose a training strategy that randomly generates training samples (with random data augmentations) while optimizing the CNN model parameters. We stop the training process and data generation until the training loss converges. In addition, we use a weighted binary cross-entropy loss function to solve the sample imbalance problem to better optimize the network, use multiple reasonable metrics to evaluate network performance, and carry out ablation experiments to optimally choose the model hyperparameters. Extensive field data applications show that our trained CNN can robustly recognize sferic signals from noisy time series for subsequent impedance estimation. The subsequent processing results show that our method can significantly improve S/N and effectively solve the problem of lack of energy in dead band. Compared to the traditional processing method without sferic compensation, our method can generate a smoother and more reasonable apparent resistivity-phase curves and depolarized phase tensor, correct the estimation error of sudden drop of high-frequency apparent resistivity and abnormal behavior of phase reversal, and finally better restore the real shallow subsurface resistivity structure.

eess.SP↗

arXiv2020

Finite Element Approximation of the Modified Maxwell's Stekloff Eigenvalues

The modified Maxwell's Stekloff eigenvalue problem arises recently from the inverse electromagnetic scattering theory for inhomogeneous media. This paper contains a rigorous analysis of both the eigenvalue problem and the associated source problem on Lipschitz polyhedra. A new finite element method is proposed to compute Stekloff eigenvalues. By applying the Babuska-Osborn theory, we prove an error estimate without additional regularity assumptions. Numerical results are presented for validation.

math.NA↗

arXiv2019

Convolutional Neural Network with Median Layers for Denoising Salt-and-Pepper Contaminations

We propose a deep fully convolutional neural network with a new type of layer, named median layer, to restore images contaminated by the salt-and-pepper (s&p) noise. A median layer simply performs median filtering on all feature channels. By adding this kind of layer into some widely used fully convolutional deep neural networks, we develop an end-to-end network that removes the extremely high-level s&p noise without performing any non-trivial preprocessing tasks, which is different from all the existing literature in s&p noise removal. Experiments show that inserting median layers into a simple fully-convolutional network with the L2 loss significantly boosts the signal-to-noise ratio. Quantitative comparisons testify that our network outperforms the state-of-the-art methods with a limited amount of training data. The source code has been released for public evaluation and use (https://github.com/llmpass/medianDenoise).

cs.CV↗

arXiv2019

Ground states and their characterization of spin-$F$ Bose-Einstein condensates

The computation of the ground states of spin-$F$ Bose-Einstein condensates (BECs) can be formulated as an energy minimization problem with two quadratic constraints. We discretize the energy functional and constraints using the Fourier pseudospectral schemes and view the discretized problem as an optimization problem on manifold. Three different types of retractions to the manifold are designed. They enable us to apply various optimization methods on manifold to solve the problem. Specifically, an adaptive regularized Newton method is used together with a cascadic multigrid technique to accelerate the convergence. According to our limited knowledege, our method is the first applicable algorithm for BECs with an arbitrary integer spin, including the complicated spin-3 BECs. Extensive numerical results on ground states of spin-1, spin-2 and spin-3 BECs with diverse interaction and optical lattice potential in one/two/three dimensions are reported to show the efficiency of our method and to demonstrate some interesting physical phenomena.

math.NA↗

arXiv2016

A Multilevel Correction Method for Interior Transmission Eigenvalue Problem

In this paper, we give a numerical analysis for the transmission eigenvalue problem by the finite element method. A type of multilevel correction method is proposed to solve the transmission eigenvalue problem. The multilevel correction method can transform the transmission eigenvalue solving in the finest finite element space to a sequence of linear problems and some transmission eigenvalue solving in a very low dimensional spaces. Since the main computational work is to solve the sequence of linear problems, the multilevel correction method improves the overfull efficiency of the transmission eigenvalue solving. Some numerical examples are provided to validate the theoretical results and the efficiency of the proposed numerical scheme.

math.NA↗

arXiv2015

A Regularized Newton Method for Computing Ground States of Bose-Einstein condensates

In this paper, we propose a regularized Newton method for computing ground states of Bose-Einstein condensates (BECs), which can be formulated as an energy minimization problem with a spherical constraint. The energy functional and constraint are discretized by either the finite difference, or sine or Fourier pseudospectral discretization schemes and thus the original infinite dimensional nonconvex minimization problem is approximated by a finite dimensional constrained nonconvex minimization problem. Then an initial solution is first constructed by using a feasible gradient type method, which is an explicit scheme and maintains the spherical constraint automatically. To accelerate the convergence of the gradient type method, we approximate the energy functional by its second-order Taylor expansion with a regularized term at each Newton iteration and adopt a cascadic multigrid technique for selecting initial data. It leads to a standard trust-region subproblem and we solve it again by the feasible gradient type method. The convergence of the regularized Newton method is established by adjusting the regularization parameter as the standard trust-region strategy. Extensive numerical experiments on challenging examples, including a BEC in three dimensions with an optical lattice potential and rotating BECs in two dimensions with rapid rotation and strongly repulsive interaction, show that our method is efficient, accurate and robust.

math.NA↗