EAGE London: Democratizing large-scale inverse problem with PyLops

The talk will have a hybrid form: in-person at Imperial College, London and broadcasted ONLINE.

Link to the webinar will be provided via e-mails to registered attendees: first e-mail will be sent two days before the event and the second one just 2 hours before the event. Do not register too late!

Lecture Theatre 2.28

Agenda (UK time)

18:30-18:35 Introduction and Announcements

18:35-19:20 Democratizing large-scale inverse problem with PyLops

19:20-19:35 Q&A

from 20:00 Informal discussion and networking

Presenters

Matteo Ravasi is a Senior Research Advisor and AI/MLOps Engineer at Shearwater Geoservices. Prior to that, Matteo was an Assistant Professor at KAUST in the School of Earth Science and Engineering, member of the Extreme Computing Research Center, and co-Director of the DeepWave industry funded consortium. Matteo also worked in Equinor both within research and operations and has led the development of several open-source software products in the geophysical domain. He holds a Phd in Geophysics from the University of Edinburgh and an MSc and BSc in Telecommunication Engineering from Politecnico di Milano. Matteo has made several contributions in the areas of seismic processing and imaging by developing novel methods to improve the quality and resolution of subsurface imaging products. For his work, Matteo is the recipient of the SEG Karcher Award, EAGE Arie van Weelden award, RAS Keith Runcorn Prize, and Gustavo Sclocchi Theses Award. Matteo is also involved in the development of the open-source project PyLops, a Python framework for large-scale inverse problems on heterogeneous architectures.

Talk outline

Inverse problems are ubiquitous in science. Nevertheless, being able to solve such problems in an efficient and scalable manner requires a mix of expertise in mathematics, computer science, and domain knowledge. PyLops is a python library developed with the goal of democratizing inverse problems, allowing domain experts to focus on answering scientific questions; it provides a large variety of linear operators implemented in a matrix-free fashion with a dual backend (numpy for cpu- and cupy for gpu-based codes), functionalities to seamlessly combine them, and state-of-the art solvers. Moreover, owing to its modular design, expert users can easily integrate PyLops operators with PyProximal’s solvers, parallel computing functionalities offered by PyLops-MPI, and autograd functionalities of Pytorch and JAX. In this talk, I will provide an overview of the PyLops ecosystem and discuss the four level of parallelism that can be harnessed to easily convert any PyLops code initially written to solve small problems on a laptop to run on heterogenous CPU-GPU supercomputers.