What is a GeoHack?
A two-day coding, problem solving and social hackathon event as part of the Second EAGE Subsurface Intelligence Workshop.
When will it take place?
28-29 October 2022
Where?
Manama, Bahrain (same venue as workshop)
Do I need to register for the hackathon?
Participation is included in the workshop registration, no additional registration is required.
Here's what you can expect...
Focusing on solving big data geoscience problems in the energy industry, the GeoHack is a highly anticipated hackathon, bringing together machine learning enthusiasts, students, geoscientists, and industry specialists to exchange ideas and develop solutions to the many complex earth imaging problems the industry is faced with.
The GeoHack will facilitate a suitable atmosphere where software developers, engineers and geoscientists will spend intensive hours to hack, test and experiment with the latest advancements in machine learning algorithms against subsurface data such as seismic, logs, cores, to solve data mining, geological interpretation and quantitative reservoir characterization problems. It is a great opportunity for companies to benchmark machine learning algorithms performance, expose the most promising technologies, and understand successes and pitfalls in machine learning.
The driving force behind the GeoHack:
- We want to build the machine learning community focused on solving geoscience problems
- We want to think out-of-the-box! Solutions may be inspired by processes from outside of the oil & gas industry.
- We encourage diversity in teams! Mentors, experts, novices, geoscientists, students are very welcome to offer and test new ideas.
- We love open source software and open source data!
- We encourage industry collaboration, entrepreneurship and innovative startups
Image credit: Sadaqat Ali Syed
Hackathon Breakdown
Day 1
Lectures
Bring everyone on the same level of ML understanding
Brief introduction to chosen seismic tasks and data format
Materials for e-learning available for participants upon registration
Practice
Guide students with “pre-filled” project in Python, help launch first model on server
Online participants follow guidelines and submit questions via group chat
Project
After successful demo-case proceed with data release and help build first approximated solutions on sight with instructors and via group chat for online
Day 2
Lectures
Explain what else is now possible in O&G with ML
Brief overview of submissions from Day 1 and online chat
Presentation of test data and leaderboard
Practice
Instructors check models and explain last implementation ideas
Students configure final submission
Result
Instructors present best solutions, demonstrate complete datasets and its tricks
Wrap up and closing
Overview
The goal of this Hackathon is to offer researchers an opportunity to develop and test various algorithms to solve routine geological tasks.
We invite you to participate in this educational event, where you get everything a data scientist can wish for: clean labeled data from multiple actual fields, pre-made Jupiter notebooks and instructors who will guide you through each step of enhancing your ML model.
Using open source libraries you will create your own model to solve Seismic Processing and Interpretation Problems.
If you have very little of Python practice, join our introductory task: First Break Picking - determine as accurately as possible onsets of first arrivals for each seismic trace in pre-stack data.
For advanced partitioners we offer a complex and challenging task: Horizon picking - determine a set of distinctly visible boundaries on seismic data: they are called horizons, and denote the drastic change in rock properties underneath.
We encourage experts in geology, math and programming to develop ML algorithms using labeled train dataset and compete with other experts by submitting results of their models on Test dataset. Those results will be evaluated using a task-specific metric and listed first on public and later on private leaderboards.
All details and demo-data will be published on GitHub page of the hackathon. Ready-to-use models, viewers and tools will be released by our instructors team! All learning materials will be available for participants after the completion of the event.
Schedule
Day 1
- Introduction to Exploration stage of Seismic Interpretation (30 min)
- Introduction to Neural Networks and Open Machine Learning Tools (30 min)
- Quick success project: demo Jupiter Notebook launch in the cloud with instructors (1 hour). Building first ML model
- GeoHack Task Intro: Horizon Detection and Data Release (1 hour)
- Team work with instructors - 1 hour
- Rest of the day - self-guided model development (Telegram chat support by instructors)
Day 2
- Recap of Day 1 (30 min)
- Team work with instructors (1 hour)
- Lecture: Major Data Science breakthrough review in Oil & Gas (1 hour)
- Final submission
- Grading of participants (1 hour)
- Closing session and prize-winners announcement (30 min)