Ausar Geophysical

Seismic processing R&D

[email protected]

Open source projects


By implementing forward modelling and backpropagation for the wave equation and Born modelling, on CPUs and GPUs, as a PyTorch module, Deepwave allows you to treat wave propagation as a step in a differentiable graph and benefit from the ease-of-use and automatic differentiation capabilities of PyTorch. You can easily use it for regular FWI, RTM, and LSRTM, but you can also add new steps before or after wave propagation, such as a deep neural network that generates the velocity model or a fancy (differentiable) cost function, and let PyTorch run forward modelling and backpropagation end-to-end.

View on GitHubDocumentation

Recent Research

Active learning for seismic processing parameterisation, with an application to first break picking

Applying active learning, a branch of machine learning that attempts to optimally select samples for labelling, to first break picking. This is an example of using machine learning to assist in labour-intensive seismic processing tasks without resorting to a black box deep neural network.


Seismic Data Denoising and Deblending Using Deep Learning

A U-net, constructed with a pretrained ResNet, incorporating information from neighbouring gathers, and trained using a diverse synthetic dataset, can denoise and deblend real datasets from different parts of the world.


Generative Adversarial Networks for Model Order Reduction in Seismic Full-Waveform Inversion

Training a Generative Adversarial Network to produce realistic seismic wave speed models, and integrating the generator network into seismic Full-Waveform Inversion, reduces the number of model parameters and restricts the inverted models to only those that are plausible.


Seismic Full-Waveform Inversion Using Deep Learning Tools and Techniques

Demonstration that the conventional seismic Full-Waveform Inversion algorithm can be constructed as a recurrent neural network and so implemented using deep learning software such as TensorFlow.