Tuesday, 5 May, 2025 / Edition 57

Has the AI fatigue set in yet? Not too much. A lot of folks are looking for the TLDR version of how to learn, what to implement, and how to use the tech.

For those folks, I have good news and bad news.

The bad news: There is no universal solution for learning or implementing AI that can be boiled down into an easy-to-read snippet. Implementation, and the skillsets needed to implement successfully, are areas in which only you/your employers know exactly what you need.

The good news: This is a YOU-directed journey. Once you know what you need, you can seek out ways to build those skillsets and tools specifically—which narrows down your reading.

For this week at least, I have found a relatively simple rundown of neural networks. Let’s take a look!

Sarah Compton

Editor, Enspired

Neural Network Basics

Everything Possible/Shutterstock.com

Neural networks are all the buzz, and many of us were introduced to them through either Petrel or some other software package where we pretended to totally understand what it was doing to spit out a fancy result.

A recent article on HackerNews.com summarized neural networks succinctly and accurately. It says...

The basic idea of neural networks is that you have training data, which includes:

• X, the input data

• Y, the output value you are trying to predict

Since we don’t know exactly how X is related to Y, we can throw together n number of general equations (“layers”), such as:

• a1 = j(W1*x + b1)

• a2 = j(W2*a1 + b2)

• an = j(Wn*a(n-1) + bn…. and so on.

In those equations, W is a weighting, b is a bias, and j is some kind of nonlinear “activation” function.

• An “activation” function determines the output of a neuron (i.e. those mathematical functions) in a neural network by helping the neural network use important information, while suppressing irrelevant data points.

The equations predict Y based on inputs for X, and hopefully, that prediction is close enough to Y to be useful.

When to use neural networks: A great example scenario for when to use a neural network would be if you want to know if you should go to an event. Let’s run through that:

• You’d consider the weather, who is attending, and the time of the event.

• The function to represent weather could be weighted heavily if you care a lot about the weather, or if it impacts your ability to travel. The timing of the event might have a similar weight.

• The functions used to represent attendees might be weighted less than weather and time.

• You can set up the network so that you run through various scenarios: Maybe it’s raining, but your bestie will be there, and it’s at a perfect time. Or maybe the weather and time is great, but the attendee list isn’t so hot.

That example is very simplified, but it gives an idea of the what and how of using neural networks.

Industry example: Geophysicists have really taken off with neural networks, using them to identify the primary reflections in their data, processing, characterization, and computing travel times for a complete 3D volume model, among other tasks.

Why it matters: Neural networks are most useful for solving hard-to-model problems analytically and can make use of the huge volumes of data available to the industry to help make better decisions.

For more information on activation functions, go here; look here for a free book on neural networks; and read here for a solid paper on neural networks in petroleum.

Sponsored

Power Your Exploration with TRITON 2024.2

Access more data, streamline planning, and analyze faster with real-time tools, ArcGIS integration, and modernized workflows.

Animal Meets Machine in this Hydrogen-Powered Motorcycle

Watch this video on Kawasaki's latest hydrogen-powered motorcycle/ Kambalitos Design and Print 

The geosciences have always struggled with accessibility in field work: Can we train good geologists without getting them into the field, and if not, are we automatically excluding folks who cannot hike their way to the outcrops?

There have been a few developments in drone technology, and some wheelchairs are being built with tracks rather than wheels, but solid solutions are still lacking, and it might be that we are too restricted in our thinking.

Taking old inspiration: Land travel used to involve riding a four-legged critter—usually a horse or camel—and while those modes of transportation are slower than motorized travel, their lack of wheels can make them nimbler and allow for travel into more rugged wilderness.

Driving the news: Kawasaki, the company I associate with motorcycles, has come up with a vehicle inspired by those kinds of critters, and they threw a 150-cc hydrogen engine in there to power it. Picture a wolf, horse, or lion-meets-motorcycle robot.

Their design includes a few eye-catching features:

• To me, the look mimics a wolf rather than a horse or camel, and I’m wondering if balance and weight distribution play a role here since a wolf allows for a lower center of gravity and shorter legs.

• The machine is controlled by the rider’s weight shifts, and AI keeps it balanced and adapts its movements in real time.

• There’s a display showing how much fuel is left, navigation tips, and weight distribution details.

• Each robot leg has rubber hooves that grip different surfaces securely.

• My time Jeeping (essentially, four-wheeling in a Jeep) has taught me that a set of slightly deflated set of sticky, 40-inch tires combined with a solid Dana 60 axle will get you A LOT of places, but man…idk if anything can compete with a set of hooves.

The vehicle is called Corleo, a derivation of Cor Leonis, the brightest star in the constellation Leo.

Kawasaki sees Corleo as an alternative to traditional motorcycles or ATVs, something that could make exploring tough terrains safer, easier, and more sustainable, but we’ll have to wait—they estimate it’s two and a half decades away from production.

Still, maybe one day we will ride a motorized, hooved animal/vehicle to the top of outcrops for field work!

👍 If you enjoyed this edition of Enspired, consider supporting AAPG's brand of newsletters by forwarding to a friend or colleague and signing up for our other newsletters here.

➡️ Was this email forwarded to you? Sign up for Enspired here.

✉️ To get in touch with Sarah or suggest a topic for Enspired, send an email to editorial@aapg.org.