Seismic waveform modeling is a powerful tool for determining earth structure models and unraveling earthquake rupture processes, but it usually computationally expensive. We introduce scheme to vastly accelerate these calculations with recently developed machine learning paradigm called the neural operator. Once trained, can simulate full wavefield at negligible cost. use U-shaped operator learn general solution 2D elastic wave equation from an ensemble of numerical simulations performed...

Abstract Seismic wave propagation forms the basis for most aspects of seismological research, yet solving equation is a major computational burden that inhibits progress research. This exacerbated by fact new simulations must be performed whenever velocity structure or source location perturbed. Here, we explore prototype framework learning general solutions using recently developed machine paradigm called neural operator. A trained operator can compute solution in negligible time any...

We consider solving ill-posed imaging inverse problems without access to an explicit image prior or ground-truth examples. An overarching challenge in is that there are many undesired images fit the observed measurements, thus requiring priors constrain space of possible solutions more plausible reconstructions. However, applications it difficult potentially impossible obtain learn prior. Thus, inaccurate often used, which inevitably result biased solutions. Rather than problem using encode...

High quality black hole videos can provide key evidence of astrophysical processes that single static images cannot provide. However, reconstructing a video is highly ill-posed problem, requiring additional structural constraints to produce plausible solution. Traditional on the spatial or temporal structure are subject human bias. In our work, we adapt recently developed techniques solve realistic reconstruction without direct priors structure, mitigating particular, set per-frame imaging...

Regression on function spaces is typically limited to models with Gaussian process priors. We introduce the notion of universal functional regression, in which we aim learn a prior distribution over non-Gaussian that remains mathematically tractable for regression. To do this, develop Neural Operator Flows (OpFlow), an infinite-dimensional extension normalizing flows. OpFlow invertible operator maps (potentially unknown) data space into process, allowing exact likelihood estimation point...

We consider solving ill-posed imaging inverse problems without access to an image prior or ground-truth examples. An overarching challenge in these is that infinite number of images, including many are implausible, consistent with the observed measurements. Thus, priors required reduce space possible solutions more desirable reconstructions. However, applications it difficult potentially impossible obtain example images construct prior. Hence inaccurate often used, which inevitably result...

Seismic waveform modeling is a powerful tool for determining earth structure models and unraveling earthquake rupture processes, but it usually computationally expensive. We introduce scheme to vastly accelerate these calculations with recently developed machine learning paradigm called the neural operator. Once trained, can simulate full wavefield at negligible cost. use U-shaped operator learn general solution 2D elastic wave equation from an ensemble of numerical simulations performed...

We consider solving ill-posed imaging inverse problems without access to an explicit image prior or ground-truth examples. An overarching challenge in is that there are many undesired images fit the observed measurements, thus requiring priors constrain space of possible solutions more plausible reconstructions. However, applications it difficult potentially impossible obtain learn prior. Thus, inaccurate often used, which inevitably result biased solutions. Rather than problem using encode...

We consider solving ill-posed imaging inverse problems without access to an image prior or ground-truth examples. An overarching challenge in these is that infinite number of images, including many are implausible, consistent with the observed measurements. Thus, priors required reduce space possible solutions more desirable reconstructions. However, applications it difficult potentially impossible obtain example images construct prior. Hence inaccurate often used, which inevitably result...

Seismic wave propagation forms the basis for most aspects of seismological research, yet solving equation is a major computational burden that inhibits progress research. This exacerbated by fact new simulations must be performed when velocity structure or source location perturbed. Here, we explore prototype framework learning general solutions using recently developed machine paradigm called Neural Operator. A trained Operator can compute solution in negligible time any location. We...

The application of infrared hyperspectral imagery to geological problems is becoming more popular as data become accessible and cost-effective. Clustering classifying spectrally similar materials often a first step in applications ranging from economic mineral exploration on Earth planetary Mars. Semi-manual classification guided by expertly developed spectral parameters can be time consuming biased, while supervised methods require abundant labeled difficult generalize. Here we develop...