Developing Deep Learning Applications for the Oilfield: From Theory to Real World Projects | 1-5 March 2021

Brief course description:
This course starts with the story of AI from the 1950s to the AI revolution of the 2010s. Today’s AI business landscape and what’s happening in the oil and gas are reviewed. The basic concepts of Machine Learning are introduced, followed by a detailed presentation of neural networks (NN). NN used 25 years ago in the oilfield are discussed, then participants experiment a NN trained in 12 seconds to read handwritten digits. Deep convolutional NNs (DNN) are then studied and applications to salt mapping and fault detection in seismic data discussed and demo’ed. Finally, participant teams are asked to design DNNs and to define how to create the data sets for two oilfield projects: sweet spot detection in Bakken field, USA, and GWC prediction in seismic data of undrilled carbonate pinnacles offshore Central Luconia, Malaysia.

Course objectives:
Upon completion of this course, participants will have acquired detailed knowledge of what Deep Learning is exactly, how it works, and in which way it differs from traditional neural networks that have been used in the industry during the past 30 years. They will understand which domains this can be applied to and for what type of applications. They will also be aware of the main challenges, difficulties, and pitfalls when developing new AI applications. Finally, they will have seen demos of deep neural networks applied to E&P disciplines and will be able to evaluate how useful the technology could be for their own domain.

Participant profiles:
This course is intended for technical staff and managers interested to understand what Deep Learning is and how it can be applied to oilfield projects (all disciplines).

About the instructor:
Dr. Bernard Montaron is CEO of Fraimwork SAS, Paris, France, and CTO of Cenozai Sdn Bhd, Kuala Lumpur, Malaysia. Two startups created in 2017 that are specialized in the application of AI to various domains, and provide services to oil and gas companies for exploration and production. In 2015-2017 he was Chief Geoscientist of Beicip Tecsol in Kuala Lumpur. Prior to this, Bernard Montaron worked 30 years for Schlumberger where he held a number of positions in R&D and Marketing. He has worked for the oil and gas industry in Europe, in the US, in the Middle East, in China, and Malaysia. Bernard was General Manager of the Schlumberger Riboud Product Center in Paris – Clamart, France (2002-2003) and he was VP Marketing of Schlumberger Middle East and Europe-Africa-Russia regions (2000-2001). Bernard holds a MSc degree in physics from ESPCI, Paris, and a PhD in Mathematics from University Pierre et Marie Curie, Paris. He has a Machine Learning certificate from Andrew Ng’s course (Stanford Univ./Coursera). Bernard received the best oral presentation award at the APGCE 2017 conference for his paper on “Deep Learning Technology for Pattern Recognition in Seismic Data – A Practical Approach”.

Course outline:

Part 1:

• A little history of artificial intelligence (A.I.)

• The AI revolution in the last ten years

• Today’s AI business landscape

• GAFAMs and free online AI software development tools

• What’s happening with AI in the Oil & Gas business?

• Linear Regression as an introduction to Machine Learning

Part 2:

• A little more on linear regression

• Neural Network basics

• Logic gates (NOT, OR, AND) as neural networks… and implications

• Understanding ML buzz words:  gradient descent, backpropagation, generalization, etc.

• The mathematics of NNs and ML (linear algebra)

• Image classification/pattern recognition with a neural network  (NN-DIGITS)

• NN-DIGITS training demo

Part 3:

• Discussion on NN from the past in the oilfield industry, and handwritten digit classification

• The Deep Learning revolution:  how did it happen?

• A very complex NN: the human eye and the visual cortex

• Introduction to deep convolutional NNs  (CNNs)

• How CNNs make it possible to develop a micro-paleontology ML software

• Free download AI tools (ATOM, Keras, TensorFlow) running on a gamer laptop PC

• NN/CNN demos with these tools using an online data set from Kaggle  

Part 4:

• Discussion on self-driving cars ML software, and adversarial attacks

• Adversarial examples for the human brain: optical illusions and perception errors

• Discussion on the salt classification paper

• Important considerations on accuracy metric

• Review of a CNN-based system for picking faults in seismic data

• Software demo of “MyFault”

• Introduction to part-5 teamwork exercises:  do your own ML project

Part 5:

• Team exercises and presentations