Dataset Information

An empirical study of large-scale data-driven full waveform inversion.

ABSTRACT: This paper investigates the impact of big data on deep learning models to help solve the full waveform inversion (FWI) problem. While it is well known that big data can boost the performance of deep learning models in many tasks, its effectiveness has not been validated for FWI. To address this gap, we present an empirical study that investigates how deep learning models in FWI behave when trained on OPENFWI, a collection of large-scale, multi-structural, synthetic datasets published recently. In particular, we train and evaluate the FWI models on a combination of 10 2D subsets in OPENFWI that contain 470 K pairs of seismic data and velocity maps in total. Our experiments demonstrate that training on the combined dataset yields an average improvement of 13.03% in MAE, 7.19% in MSE and 1.87% in SSIM compared to each split dataset, and an average improvement of 28.60%, 21.55% and 8.22% in the leave-one-out generalization test. We further demonstrate that model capacity needs to scale in accordance with data size for optimal improvement, where our largest model yields an average improvement of 20.06%, 13.39% and 0.72% compared to the smallest one.

SUBMITTER: Jin P

PROVIDER: S-EPMC11358280 | biostudies-literature | 2024 Aug

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

An empirical study of large-scale data-driven full waveform inversion.

Jin Peng P Feng Yinan Y Feng Shihang S Wang Hanchen H Chen Yinpeng Y Consolvo Benjamin B Liu Zicheng Z Lin Youzuo Y

Scientific reports 20240828 1

This paper investigates the impact of big data on deep learning models to help solve the full waveform inversion (FWI) problem. While it is well known that big data can boost the performance of deep learning models in many tasks, its effectiveness has not been validated for FWI. To address this gap, we present an empirical study that investigates how deep learning models in FWI behave when trained on OPENFWI, a collection of large-scale, multi-structural, synthetic datasets published recently. I ...[more]

PMID: 39198496

Similar Datasets

Project description:Studies of the apportionment of human genetic variation have long established that most human variation is within population groups and that the additional variation between population groups is small but greatest when comparing different continental populations. These studies often used Wright's F(ST) that apportions the standardized variance in allele frequencies within and between population groups. Because local adaptations increase population differentiation, high-F(ST) may be found at closely linked loci under selection and used to identify genes undergoing directional or heterotic selection. We re-examined these processes using HapMap data. We analyzed 3 million SNPs on 602 samples from eight worldwide populations and a consensus subset of 1 million SNPs found in all populations. We identified four major features of the data: First, a hierarchically F(ST) analysis showed that only a paucity (12%) of the total genetic variation is distributed between continental populations and even a lesser genetic variation (1%) is found between intra-continental populations. Second, the global F(ST) distribution closely follows an exponential distribution. Third, although the overall F(ST) distribution is similarly shaped (inverse J), F(ST) distributions varies markedly by allele frequency when divided into non-overlapping groups by allele frequency range. Because the mean allele frequency is a crude indicator of allele age, these distributions mark the time-dependent change in genetic differentiation. Finally, the change in mean-F(ST) of these groups is linear in allele frequency. These results suggest that investigating the extremes of the F(ST) distribution for each allele frequency group is more efficient for detecting selection. Consequently, we demonstrate that such extreme SNPs are more clustered along the chromosomes than expected from linkage disequilibrium for each allele frequency group. These genomic regions are therefore likely candidates for natural selection.

Dataset Information

An empirical study of large-scale data-driven full waveform inversion.

Publications

An empirical study of large-scale data-driven full waveform inversion.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets