Wed 20th February @ noon, Rm 312.222

Abstract:

Deep learning methods have achieved great success in various areas including computer vision, speech recognition, natural language processing, robotics, bioinformatics, chemistry, finance, and many others. Contrary to classical machine learning approaches, methods based on deep learning and big data achieve high efficiency and superhuman performance in many complex tasks. When applied to geophysical data, these technologies have the potential to completely transform various problems in simulations, data processing, imaging, and interpretation. In this talk, I will present several applications of deep learning in geophysical exploration and production. One of them is deep learning inversion that can provide reliable estimates of the subsurface properties orders of magnitude faster than conventional techniques. In relatively simple settings, deep neural networks trained even on relatively small datasets can reliably estimate the unknown parameter distribution with high precision. In complex geological settings, the networks can provide instantaneous estimates of the initial distribution of formation parameters to assist in fast decision-making or be used as a starting model for inversion. For some forward modelling applications, deep neural networks can approximate physical simulations with a high degree of accuracy and in orders of magnitude less time. In particular, a combination of spatial and sequence neural networks can be highly efficient in forecasting various dynamic processes. The above-mentioned problems are only a tiny fraction of the full spectrum of problems in geophysics and geology where artificial intelligence can potentially make a significant contribution. In the last part of the talk, I will discuss the recent progress in deep learning for cutting-edge areas such as robotics and video games.

Short bio:

Vladimir Puzyrev is a Senior Research Fellow at the Curtin University Oil and Gas Innovation Centre and the School of Earth and Planetary Sciences. He joined Curtin University in 2016 as a Research Fellow. Prior to taking this position, Vladimir worked at Barcelona Supercomputing Center on the development of a high-performance framework for inversion of large-scale geophysical data. His current research focuses on applications of artificial intelligence and deep learning in geosciences. His research interests also include numerical methods for PDEs, computational electromagnetics and high-performance computing.