Common Source Domain Research Articles

Gradient Matching Federated Domain Adaptation for Brain Image Classification.

Federated learning has shown its unique advantages in many different tasks, including brain image analysis. It provides a new way to train deep learning models while protecting the privacy of medical image data from multiple sites. However, previous studies suggest that domain shift across different sites may influence the performance of federated models. As a solution, we propose a gradient matching federated domain adaptation (GM-FedDA) method for brain image classification, aiming to reduce domain discrepancy with the assistance of a public image dataset and train robust local federated models for target sites. It mainly includes two stages: 1) pretraining stage; we propose a one-common-source adversarial domain adaptation (OCS-ADA) strategy, i.e., adopting ADA with gradient matching loss to pretrain encoders for reducing domain shift at each target site (private data) with the assistance of a common source domain (public data) and 2) fine-tuning stage; we develop a gradient matching federated (GM-Fed) fine-tuning method for updating local federated models pretrained with the OCS-ADA strategy, i.e., pushing the optimization direction of a local federated model toward its specific local minimum by minimizing gradient matching loss between sites. Using fully connected networks as local models, we validate our method with the diagnostic classification tasks of schizophrenia and major depressive disorder based on multisite resting-state functional MRI (fMRI), respectively. Results show that the proposed GM-FedDA method outperforms other commonly used methods, suggesting the potential of our method in brain imaging analysis and other fields, which need to utilize multisite data while preserving data privacy.

Corpus-Based Diachronic Study of WAR Metaphor in Indonesian Political Discourse

The use of metaphor in political discourse is pervasive. One of the common source domain used in political metaphor is of WAR source domain. The increasing use of WAR metaphor reflects the intensity of the political situation. This paper examines the diachronic used of WAR metaphor in Indonesian Political Discourse. The aim is to analyze the use of the WAR source domain of metaphor used to speak about the Indonesian Political Situation and to compare the use of the metaphor before and after the direct election system is in effect (in 2004). The data are taken from 1896 news articles related to the political situation dan political coalition, that spans from 1980 to 2020 in Tempo online magazine. The study is corpus-based with The Conceptual Metaphor Framework of Lakoff and Johnson used as the theoretical background. The data analysis is conducted in two stages, first by quantitative analysis using Antconc Corpus Concordance Software to measure the frequency of WAR metaphor used to speak about Political situation and the second is the qualitative analysis to evaluate the change and the development of metaphor use. The result of the study should reflect how the political change that occurred in Indonesia can be reflected through the language in particular reflected from the metaphor used to speak about the political condition. The finding shows that there are three categories of diachronic change of WAR metaphorical expression: the first is the terms that have been used metaphorically even before 1998 (reformation era); the second is the terms that are recorded in use as a metaphor after 1998 and the last is the metaphorical expressions that are recorded in use before 1998 but have not been recorded after that.

Domain Adaptive and Generalizable Network Architectures and Training Strategies for Semantic Image Segmentation.

Unsupervised domain adaptation (UDA) and domain generalization (DG) enable machine learning models trained on a source domain to perform well on unlabeled or even unseen target domains. As previous UDA&DG semantic segmentation methods are mostly based on outdated networks, we benchmark more recent architectures, reveal the potential of Transformers, and design the DAFormer network tailored for UDA&DG. It is enabled by three training strategies to avoid overfitting to the source domain: While (1) Rare Class Sampling mitigates the bias toward common source domain classes, (2) a Thing-Class ImageNet Feature Distance and (3) a learning rate warmup promote feature transfer from ImageNet pretraining. As UDA&DG are usually GPU memory intensive, most previous methods downscale or crop images. However, low-resolution predictions often fail to preserve fine details while models trained with cropped images fall short in capturing long-range, domain-robust context information. Therefore, we propose HRDA, a multi-resolution framework for UDA&DG, that combines the strengths of small high-resolution crops to preserve fine segmentation details and large low-resolution crops to capture long-range context dependencies with a learned scale attention. DAFormer and HRDA significantly improve the state-of-the-art UDA&DG by more than 10 mIoU on 5 different benchmarks.

Metaphors in Jason Mraz's Song Lyrics (An Ecolinguistic Perspective)

This research was intended to analyze Metaphor in Jason Mraz’s song lyrics based on an Ecolinguistic perspective by Kovecess theory. The objective of this research was “To describe how metaphors delineation realized in Jason Mraz’s song lyrics based on the common source domain and the common target domain”. This research used a descriptive qualitative method. The researcher used the metaphor theory by Kovecess to analyze the data. The source of the data in this research was taken from Jason Mraz’s song lyrics such as The World as I See it, Back to the Earth, and Wise Woman. The data was collected through downloading, reading, classifying, coding, describing, and interpreting the data to get a complete conclusion of the Metaphor in Jason Mraz’s song lyrics based on an Ecolinguistic perspective. The researcher analyzed lyrics that show metaphor based on the common source domain and the common target domain in the lyrics. The result showed that the metaphor delineations were revealed from the analysis process by using the mapping common source domain and common target domain by Kovecess. The researcher found 19 Metaphors include the common source domain and the target domain. It showed the condition of the world that has a remarkable place, the beauty of ecology, our role as a part of nature. The three songs also convey an implicit message about how protecting and loving the environment create harmony and live peacefully because humans and nature have an interrelation.
 Keywords: metaphor, ecolinguistics, kovecses, sog lyrics, source domain and target domain

Deblending of Simultaneous-Source Seismic Data Based on Deep Convolutional Neural Network

The simultaneous-source technique improves the acquisition efficiency in marine seismic exploration. A high-quality deblending procedure is an important step when implementing the simultaneous-source technique. Deblending is often implemented in a domain (other than the common-source domain) where the blending noise is incoherent. In this article, we present a denoising method based on a deep convolutional neural network (CNN) for deblending. The denoising generalization ability of CNN-based denoising on real seismic data is limited because the training dataset, especially the labeled data, is often difficult to acquire. To make the CNN applicable to real data, we generate the training dataset directly from the real common-shot blended record itself, which has the same dynamic seismic wavefield characteristics as the real data. The input dataset for the CNN is acquired by adding A to B. A is the random time-delay encoding data of each trace of the common-shot blended data, while B is the data of each trace of the common-shot blended data; then, the blended data naturally become labeled data. The CNN model obtained through deep learning is used to remove the blending noise to complete the separation of blended data. Numerical tests using synthetic data and real data show that our method can provide high-precision separation results.

Adjoint-Driven Deep-Learning Seismic Full-Waveform Inversion

Seismic full-waveform inversion (FWI) aims to build high-resolution images of the physical properties of the subsurface. However, the ill-posedness and nonlinear problems pose a great challenge to the high-resolution reconstruction. Although the nonlinear problem can be mitigated by matching a subset of observation data, the resulting images are generally low-resolution background structures. Regularization-based techniques can mitigate the ill-posedness of FWI, but the iterative method suffers from the cycle-skipping and computational burden problems. To overcome these problems, we develop an adjoint-driven deep-learning FWI (AD-DLFWI) approach which utilizes the fully convolutional network (FCN) to invert subsurface velocity from reflection seismic data. Specifically, AD-DLFWI is implemented in a two-step iterative scheme, in which an optimal update result at each step is learned via a FCN-based learned updating operator. The proposed approach uses the seismic image of applying the adjoint operator of the scattering wave equation, which is equivalent to the gradient of classical FWI, as the data engine of FCN. Inspired by the wave-equation migration velocity analysis approach, we propose to unfold the gradient of FWI into the common-source domain to keep the information about the measure of velocity error. To ensure the interpretability of each network’s role, we design a two-step training scheme to successively reconstruct the low and high wavenumber components of subsurface velocity. Using synthetic experiments with reflection-dominant seismic data, we have confirmed that the proposed FWI approach not only can provide a reliable velocity estimation but also is not sensitive to the cycle-skipping problem.

Normalized shaping regularization for robust separation of blended data

Simultaneous source acquisition has attracted more and more attention from geophysicists because of its cost savings, whereas it also brings some challenges that have never been addressed before. Deblending of simultaneous source data is usually considered as an underdetermined inverse problem, which can be effectively solved with a least-squares (LS) iterative procedure between data consistency ([Formula: see text]-norm) and regularization ([Formula: see text]-norm or [Formula: see text]-norm). However, when it comes to abnormal noise that follows non-Gaussian distribution and possesses high-amplitude features (e.g., erratic noise, swell noise, and power line noise), the [Formula: see text]-norm is a nonrobust statistic that can easily lead to suboptimal deblended results. Although abnormal noise can be attenuated in the common source domain at first, it is still challenging to apply a coherency-based filter due to the sparse receiver or crossline sampling, e.g., that commonly found in ocean bottom node (OBN) acquisition. To address this problem, we have developed a normalized shaping regularization to make the inversion-based deblending approach robust for the separation of blended data when abnormal noise exists. Its robustness comes from the normalized shaping operator defined by the confidence interval of normal distribution, which minimizes the abnormal risk to a normal level to satisfy the assumption of LS shaping regularization. In special cases, the proposed approach will revert to the classic LS shaping regularization once the normalized coefficient is large enough. Experimental results on synthetic and field data indicate that the proposed method can effectively restore the separated records from blended data at essentially the same convergence rate as the LS shaping regularization for the abnormal noise-free scenario, but it can obtain better deblending performance and less energy leakage when abnormal noise exists.

CONTRASTIVE ANALYSIS OF EXPRESSIONS ON JAPANESE AND INDONESIAN LOVE LYRICS -BASED ON COGNITIVE LINGUISTIC POINT OF VIEW-

Song is an expression which has a strong connection with someone's feeling, which can also be a hint to understand how Japanese society thinks and feels in general (Kanemoto 2006). Expression on song lyrics is quite different from the usual expression used in daily conversation. To convey emotions and feelings of the songwriter, the style of language is important to touch the listener's feelings. This research analyzed the style of language in the lyrics of Japanese and Indonesian love song, by using contrastive analysis method and review it from cognitive linguistics. 13 Common Source Domains that Kovecses exposes is used to identify the style of love expression in the song lyrics. The purpose of this study is to explain what language styles and expression are used in the lyrics of Indonesian and Japanese songs, accordance with theory of the seven metaphors of love in cognitive linguistics, contrasting the love phrases contained in both languages, and the corresponding theory of the seven metaphors of love in cognitive linguistics with the Common Source Domain by Kovecses. As the results, there are 8 types of 13 types of Common Source Domain used in the lyrics of both languages in the 1970s. Besides the 13 Common Source Domains, is also found domain that are used both languages called Colour. Meanwhile the song lyrics in 2000 year, there are 10 types of 13 types of Common Source Domain used in both languages lyrics. Another domain that is used in both language lyrics is Music.

Oriented time-domain dip moveout correction for planar reflectors in common-source domain

Oriented time-domain imaging can be orders of magnitude faster than the routine techniques, which rely on velocity analysis. The term “oriented” refers to those techniques that use the information carried by local slopes. Time-domain dip moveout (DMO) correction, which has often been ignored by the seismic imaging community, has been coming back to attention within the last few years. I have developed an oriented time-domain DMO correction workflow that does not face the problematic loop between the dip-dependent and/or dip-independent velocities existing in the classic DMO correction algorithms. Use of the proposed approach is also advantageous over the previous oriented techniques; the proposed technique is independent of the wavefront curvature, and the input seismic data do not need to be sorted in two different domains. The application of the technique is limited to reflectors with a small curvature. The theory of the proposed technique is investigated on a simple synthetic data example and then applied to a 2D marine data set.

Oriented prestack time migration using local slopes and predictive painting in the common-source domain for planar reflectors

Seismic imaging techniques often require an input velocity model. Velocity analysis is one of the most critical stages in seismic data processing. Standard ways to find the velocity model from seismic data in the time domain are constant velocity stack and semblance velocity analysis that may be time consuming and labor intensive. Oriented/velocity-less imaging using local event slopes is an alternative to the conventional imaging techniques. In some previous oriented techniques, seismic data must be sorted in two different domains, whereas seismic data are not always available in both domains and the use of interpolation is inevitable in such cases. Other methods are developed in terms of the higher order derivatives of traveltime with respect to offset, whereas estimation of the higher order derivatives is difficult to achieve with the required accuracy. We addressed the limitations by developing an oriented local slope based prestack time migration technique in only one domain: the common-source domain. The migration technique is developed for reflectors with small curvature. In the proposed approach, the need for the estimation of higher order derivatives is replaced by a point-to-point mapping of seismic data using the predictive painting technique. The theoretical contents of the proposed technique are tested on a simple synthetic data example and applied to a field data set.

Velocity analysis using zero-offset attributes in common source domain

SUMMARY Velocity analysis is one of the most critical stages in seismic data processing and the velocity errors influence the accuracy of the imaging. The conventional workflows, to obtain the migration velocity model, are generally labour intensive and time consuming. An experienced processor is needed to pick the velocities in the velocity spectrum. In this paper, we introduce another approach to obtain the velocity model using the first derivative of time with respect to offset (local slopes) in one domain, common source domain. In this time- effective approach, one order of differentiation has been reduced from the previous velocity analysis in common source domain, using zero-offset attributes. This velocity analysis and imaging are being done at the same time. There is no need to have any velocity model prior to imaging in this approach. The accuracy of the velocity analysis method is more robust than the previous technique because of the reduction in order of the derivatives needed. Computational experiments with synthetic seismic data examples confirm the theoretical expectations and demonstrate the feasibility of the proposed technique.

Pre-stack time migration in common source domain without velocity model

Most of the migration techniques require an input velocity model. Velocity analysis is one of the most critical stages in seismic data processing. Standard ways to find the velocity model are const...

Iterative method for the separation of blended seismic data: discussion on the algorithmic aspects

ABSTRACTRecently much attention has been given to the possibility of shooting in an overlapping fashion, the so‐called blended or simultaneous acquisition. In conventional acquisition the time intervals between successive sources are large enough to avoid interference in time. In blending, a temporal overlap between source responses is allowed. This additional degree of freedom in survey design has the potential to significantly reduce seismic acquisition costs while maintaining or improving the data quality. Deblending is the procedure of retrieving data as if they were acquired in the conventional, unblended way. This is an essential step in the case where standard processing flows are applied. Several methods have been proposed to perform data separation with the majority of them falling into two categories. The first category consists in methods that filter out the blending noise by arranging seismic data in some domain, whereas inversion techniques fall into the second category. We recently introduced an iterative estimation and subtraction algorithm that integrates elements of both categories. This method exploits the fact that the character of the blending noise differs in different domains, e.g., it is coherent in the common source domain but incoherent in the common receiver, common‐offset or common midpoint domains. Up to now, our method relies on the interaction with a human operator. The automation of the thresholding process is addressed in this paper, leading to a hands‐off algorithm for the separation of blended data, optimized for both efficiency and effectiveness. We found that one of the major limiting factors is the edge artefacts generated by the coherence‐pass filter. The effectiveness of the coherence‐pass filter has a considerable influence on the convergence of our deblending algorithm. This is shown by testing different coherence‐pass filters on marine data as well as introducing errors in the coherence‐pass process. We show that these data estimation errors can be handled properly and the best result is obtained using a τ‐p filter as a coherence‐pass filter. Furthermore, very promising results are obtained on numerically‐blended land data.

Separation of blended data by iterative estimation and subtraction of blending interference noise

Seismic acquisition is a trade-off between economy and quality. In conventional acquisition the time intervals between successive records are large enough to avoid interference in time. To obtain an efficient survey, the spatial source sampling is therefore often (too) large. However, in blending, or simultaneous acquisition, temporal overlap between shot records is allowed. This additional degree of freedom in survey design significantly improves the quality or the economics or both. Deblending is the procedure of recovering the data as if they were acquired in the conventional, unblended way. A simple least-squares procedure, however, does not remove the interference due to other sources, or blending noise. Fortunately, the character of this noise is different in different domains, e.g., it is coherent in the common source domain, but incoherent in the common receiver domain. This property is used to obtain a considerable improvement. We propose to estimate the blending noise and subtract it from the blended data. The estimate does not need to be perfect because our procedure is iterative. Starting with the least-squares deblended data, the estimate of the blending noise is obtained via the following steps: sort the data to a domain where the blending noise is incoherent; apply a noise suppression filter; apply a threshold to remove the remaining noise, ending up with (part of) the signal; compute an estimate of the blending noise from this signal. At each iteration, the threshold can be lowered and more of the signal is recovered. Promising results were obtained with a simple implementation of this method for both impulsive and vibratory sources. Undoubtedly, in the future algorithms will be developed for the direct processing of blended data. However, currently a high-quality deblending procedure is an important step allowing the application of contemporary processing flows.

A solution to the airwave-removal problem in shallow-water marine EM

A new method is presented for removing the airwave in shallow-water electromagnetic (EM) data. By using pairs of receivers or pairs of source points along a towline and forming weighted differences of an EM field component, the airwave is significantly attenuated. The weights are related to the geometric spreading of the airwave. Specifically, the horizontal electric and magnetic field components are weighted with the horizontal offset to the power of three before their differences are calculated. One can also first calculate difference data without weights in the common-source domain to remove correlated noise and then use the resulting data to perform a second-order difference with weights. In this case, the weights should equal the horizontal offset to the power of four. Moreover, the weighted difference data can be integrated to obtain an estimate of the physical field void of the airwave. Additionally, the airwave contribution to the physical field can be estimated. The theory behind the method is presented and demonstrated on a real controlled-source EM (CSEM) data set.

Metaphor awareness and vocabulary retention

Various figurative expressions can often be traced back to a common metaphoric theme or source domain. The present article reports three EFL experiments which indicate that a lexical organization along such metaphoric themes or source domains can facilitate retention of unfamiliar figurative expressions. In view of these findings, the article proposes classroom activities aimed at enhancing language learners' metaphor awareness and at turning this into an additional channel for vocabulary acquisition.