Center for Machine Learning for Seismic Industry Partners Program  

Mission

Georgia Tech’s Center for Machine Learning for Seismic (ML4Seismic) is designed to foster research partnerships aimed to drive innovations in artificial-intelligence assisted seismic imaging, interpretation, analysis, and monitoring in the cloud.

Approach

Through training and collaboration, the ML4Seismic Industry Partners Program promotes novel approaches that balance new insights from machine learning with established approaches grounded in physics and geology. Areas of interest include, but are not limited to, low-environmental impact time- lapse acquisition, data-constrained image segmentation, classification, physics-constrained machine learning, and uncertainty quantification. These research areas are well aligned with Georgia Tech’s strengths in computational/data sciences and engineering.

Participation

Participation. Participation by PARTICIPANTS in ML4Seismic is voluntary. Any donation of cash to ML4Seismic will be made as a gift to the GTF.

Any donation of cash to the GTF will be made in compliance with the Georgia Institute of Technology’s gift policy which includes, but is not limited to, the following:

  • PARTICIPANTS receive no direct benefit and requires nothing in exchange beyond a general assurance that the intent of the gift be honored in accordance with these Operating Guidelines and other related ML4Seismic agreements

  • Any donation of cash has no contractual requirements

  • Any donation of cash is irrevocable

  • There is no period of performance associated with any donation of cash

  • Financial reporting to PARTICIPANTS is not required

  • Unexpended funds do not have to be returned to PARTICIPANTS

  • Donations of cash are voluntarily given without any expectation in return

  • There is no direct transfer of IP to PARTICIPANTS. Under no conditions will any particular RESEARCH RESULTS be linked, or given the perception of being linked to a specific donation by any individual PARTICIPANT.

Donation of other resources and other forms of participation are also voluntary, and will be coordinated with ML4Seismic on an as needed and mutually agreeable basis.

Current Research Themes

1. Seismic Data Interpretation: Interpreting faults, salt domes, and other structures of interest, as well the various rock facies is of immense importance to many tasks, ranging from carbon sequestration and storage to reservoir exploration and exploitation. The automation of this interpretation via machine learning algorithms would require manual annotation of significant amounts of unlabeled seismic data by experienced interpreters, which makes it a time-consuming and laborious process. To overcome this bottleneck, we are actively working on developing weakly-,semi-, and un-supervised machine learning methods to unburden the interpreter from manually annotated data. The incorporation of uncertainty into network estimations provides more insight to decision makers about the confidence of network predictions. This work has resulted in countless impactful publications already, along with three open-source datasets for benchmarking ML algorithms for accuracy and efficiency.

Interpretation.png

2. Property Characterization: Rock property characterization of seismic volumes plays an essential role for interpretation. While rock properties can be physically measured at the wells, they have to be estimated away from well positions. The high cost of drilling wells makes training large machine learning models especially difficult, since these algorithms can easily overfit to the limited training data, doing poorly elsewhere. We are working on efficient sequence modeling workflows that incorporate the temporal and spatial aspects of data to make more robust and accurate estimations. To overcome limited training data, we are researching on semi-supervised and Transfer Learning-based methods, to help the algorithm learn prior knowledge from unlabeled data as well as labeled data in other datasets. The injection of physics-based inverse models and uncertainty leads to highly efficient, uncertainty-aware property modeling workflows able to perform robustly with limited data.

property.png

3. Imaging: We have leveraged recent developments in transformative fields of compressive sensing and machine learning (‘big data’) to drive innovations in exploration seismology. Integrating broad geophysical insights with cutting edge sampling and computational approaches has led to major improvements (see our mind map) in seismic data acquisition, modelling, imaging, and inversion. The main breakthrough of our approach is a new framework with acquisition and computational costs that are no longer determined by pessimistic sampling criteria. Instead, our costs depend on structure, e.g., sparsity or low rank, exhibited by the final inversion results and these costs will therefore no longer grow uncontrollably with the dimensionality of the inversion problem.

imaging.png

Funding Model

Funding. ML4Seismic will be funded through the voluntary donations of interested parties to the GTF. It is envisioned that additional resources may be procured by ML4Seismic, through appropriate research or other agreements if necessary and as authorized, from foundations, individuals, research institutions, and government agencies.

Charter Partner at $180,000 or more (Annual)

  1. Disseminate job announcements and facilitate recruitment of students for internship opportunities

  2. A total of five complimentary registrations for the Fall and Spring ML4Seismic meetings when held

  3. Access to information to reproduce our results made available to the public via creative commons and open source licenses

  4. Invitation of company representatives to the Industry-Student Mixer providing opportunities to engage with both students and faculty for potential research projects, student recruitment, etc.

  5. Two complimentary registrations per year for ML4Seismic professional education course/workshop (if it is held), such as for example, professional education course/workshops on best practices for machine learning in image segmentation and classification; in seismic acquisition design, in seismic imaging in the cloud, or in the fundamentals of machine learning

  6. Facilitate engagement with ML4Seismic faculty and students

  7. Facilitate company information session once a year to discuss potential new research directions and opportunities

  8. Invitation of up to 5 company representatives to attend ML4Seismic hackatons when held and during which problems on open source (labeled) seismic data are solved

  9. Access to beta testing of ML4Seismic’s open source software

  10. Recognition as a Diamond sponsor of the ML4Seismic Seminar Series

Executive Partner at $90,000 —$179,000 (Annual)

  1. Disseminate job announcements and facilitate recruitment of students for internship opportunities

  2. A total of three complimentary registrations for the Fall and Spring ML4Seismic meetings when held

  3. Access to information to reproduce our results made available to the public via creative commons and open source licenses

  4. Invitation of company representatives to the Industry-Student Mixer providing opportunities to engage with both students and faculty for potential research projects, student recruitment, etc.

  5. One complimentary registrations per year for ML4Seismic professional education course/workshop (if it is held), such as for example, professional education course/workshops on best practices for machine learning in image segmentation and classification; in seismic acquisition design, in seismic imaging in the cloud, or in the fundamentals of machine learning

  6. Facilitate engagement with ML4Seismic faculty and students

  7. Recognition as a Platinum sponsor of the ML4Seismic Seminar Series

Policy on Intellectual Property

Intellectual Property (IP). The primary objectives of ML4Seismic are in general to advance the public welfare as it relates to improved artificial-intelligence assisted seismic imaging and interpretation and innovation through a transparent, technically rigorous, multi-stakeholder process. In transacting its operations consistent with this overarching objective, ML4Seismic will rely to a great extent upon publicly available information, data, publications, prior literature, and best practices.

Policy on Data Ownership and Use

Public Data. As a general rule, it will be the general policy of ML4Seismic to seek to identify and utilize relevant, publicly available data, software, analytical tools and other information to the greatest possible extent in the RESEARCH.

Non-Public Data. During the course of ML4Seismic RESEARCH, it may be desirable to utilize data, information, software, analytical tools or other proprietary information that is the confidential property of PARTICIPANTS, or otherwise non-public. Should ML4Seismic need access to such non-public information, data, software or analytic tool, a Data Access Agreement shall be negotiated and executed, the terms of which shall be consistent with the policy, procedures and goals of ML4Seismic and GIT.

Generation and Sharing of New Data. ML4Seismic’s goal is to drive innovations in artificial- intelligence assisted seismic imaging, interpretation, analysis, and monitoring in the cloud and to publish the RESEARCH RESULTS developed during its research activities. It will be the policy of ML4Seismic to make RESEARCH RESULTS publicly available, unless circumstances dictate otherwise, such as may be required by a non-disclosure and/or Data Access Agreement entered into regarding data, information, software or analytical tools as described herein.