*North Tech Breakfast: Neural Networks 101: Math, not Magic - May 13th
Sponsored by Anadarko and Quantico Energy Solutions
Please Note Meeting Moved to Monday, May 13th
1201 Lake Robbins Drive
The Woodlands, TX 77380
NOTE: You Must be Logged in To Register
Speaker: Scotty Salamoff, Actus Veritus Geoscience, LLC
Neural network and machine learning technology have been a part of oil and gas geoscience analysis for longer than many people think. StratimagicTM is among several examples of commercial use of neural networks since late the 1990s, yet the technology has had very limited market penetration. The underutilization (or non-use) of neural networks as critical components in subsurface interpretation workflows can be explained by a combination of not appreciating how the proverbial engine of a neural network functions, misunderstanding the output it is intended to provide, and overestimating the financial cost of implementation.
In this talk, we revisit basic mathematical principals of neural networks to provide a foundation for understanding the neural network process and analyzing results. A simple 2-layer sigmoidal neural network is used as an example to demonstrate the key components of neural network architecture, and demonstrate the mathematics and programming used to create one. While cloud-based deep neural net technology such as TensorFlow are very popular for large projects, our focus is instead on the creation of a simple neural net from scratch, which allows us to examine and discuss the elements used to build it. We discuss how input layers are created, define how hidden layers are generated, and how their numbers are constrained, and how activation functions and back-propagation of calculated error provide the fuel needed for the network to continue learning.
Some parallels between human brain structure and artificial neural networks – such as what synaptic chains are, why they are important for learning, and where they fit into a neural network – are discussed to bring into context the basics of what neural nets do and how they work. While we discuss the theory and mathematics of a simple neural network, we also show how one can build a neural network in Python so as to make the final jump from curious bystanders to AI practitioners.
Speaker Biography: Scotty Salamoff, Actus Veritus Geoscience, LLC
Scotty Salamoff is the owner and principal geophysicist and data scientist at GeoTerra Technologies as well as Chief Technology Officer at Actus Veritas. Scotty applies his passion, expertise and experience developing geophysical workflows and designing, implementing, and integrating neural networks for upstream oil and gas companies to deliver data-supported interpretation products with the goal of reducing pre-drill risk and identifying complete petroleum systems from seismic data. His workflows are designed to allow data to speak, and the interpreter to arrive at possible conclusions rather than attempting to model the data in order to fit a desired outcome.
Scotty is an experienced play validation and evaluation specialist with worldwide basin exposure. In his most recent position as Senior Exploration Geophysicist at Noble Energy, he successfully delivered two high-impact frontier exploration wells, serving as the operations and post-drill evaluation geophysicist. He also designs and codes his own attribute algorithms and software plugins when no off-the-shelf solution exists.
Current professional interests include the interpretation and integration of DAS data into field optimization plans, creating and implementing targeted neural network processes into production and exploration workflows, and detailed 3D seismic attribute analysis and seismic interpretation.
Scotty is an accomplished classroom instructor and public speaker, having taught geophysical interpretation techniques at Chevron at SCM E&P Solutions. He has authored and co-authored papers about Fracture Detection from Seismic and AVO use to identify lithology. He has also presented technical talks about non-amplitude seismic attributes and how they relate and respond to hydrocarbon pore fluids and has presented several times at DHI consortium meetings.
He holds two Master of Science degrees in Geophysics and Structural Geology from Colorado State University.
Breakfast will be available starting at 7:00 AM
Technical Talk will start at 7:30
Event will end by 8:30
THANK YOU TO OUR GENEROUS SPONSORS