Text size adjustment
Hold down the Ctrl key (PC) or Cmd key (MAC) and press "+" to enlarge or "-" to reduce the text size.
From 31st of July to 16th October 2020 FORCE with help from AGILE will arrange a globally open machine learning contest. We want to find the team that is best at predicting lithology and the stratigraphy in wells in the North Sea. A curated public dataset is being prepared.
Submissions are also open to determine the team that builds the best machine learning based fault mapping algorithm for seismic data. Scoring will be carried out by structural geology domain experts
We are currently looking for interested companies who are willing to sponsor this high publicity global event.
For more information contact: firstname.lastname@example.org
The contest will be open from July 31st to October 16th 2020.
Teams from around the world will be able to access the curated well and the seismic data. Matt Hall and his team from Agile will set up a Kaggle style webpage where people can discuss issues, share ideas and score their current machine learning algorithms against the blind test data. We expect start up and intermittent press releases to relevant web pages and print media.
At the end of September a one day online seminar will be hosted to announce the winners, show and discuss the results as well as presenting other ML geoscience topics
We aim to provide a curated, qc’ed dataset of 100-150 Norwegian wells with a high quality log suite as well as training labels for lithology, porosity and core porosity. Teams can submit their trained algorithms to scoring webpage where they will be scored against the public wells and a non public hold out set. The hold out set will be published after the competition.
The three winning machine learning models and workflows have to be published in full. The final lithology predictions will also be published in full from all participants
A seismic dataset for fault detection mapping will be provided. Fault detection can be done using either manual (4 provided labeled lines) or on synthetic seismic training data. Each team can submit only one trained algorithm. Scoring will be on a blind seismic dataset with similar fidelity, structural makeup and signal/noise ratio. For comfort some images of the scoring dataset will be made available upfront.
The scoring will be undertaken by structural geologist experts. A benchmark seismic fault detection will be provided at the beginning of the competition to show the current state of the art of fault detection on seismic data.
Sponsor fund will be used to create a high quality well log and well label dataset that can serve as a reference for future competitions and research around machine learning and well log data.
To date such a dataset does not exist.
In addition the funds will be used to manage the competition and secure the necessary publicity.
Gold Sponsor: >70 000 kr
Silver Sponsor: 25 000 kr