“We are trying to enable the geoscientists to do their work better and faster.”
In this episode, we explore the use of artificial intelligence (AI) in seismic interpretation, focusing on the advantages of a data-centric approach over the traditional model-centric method. Morten Ofstad, a computer scientist, emphasizes the limitations of pre-trained “black box” deep learning models and advocates for interactive deep learning to improve interpretation accuracy. The discussion highlights VDS, a data format designed for random access and compression, and emphasizes the importance of empowering geoscientists to interact directly with AI-driven interpretation processes.
In this episode, we talk about:
- The differences between model-centric and data-centric approaches to AI in seismic interpretation.
- The limitations of “black box” deep learning models in seismic interpretation** and how an interactive approach can improve accuracy and insights.
- The importance of high-quality data and accurate labels in training AI models** for seismic interpretation and how the data-centric approach helps identify inaccuracies.
- How virtual data storage (VDS), a data format designed for random access and compression, can improve the efficiency of data-centric AI workflows** in seismic interpretation.
- The potential of data-centric AI to empower geoscientists**, enabling them to work faster and more accurately.
Sponsor
Bluware’s InteractivAI is a human-powered AI seismic analysis tool, revolutionizing the way geoscientists extract value from seismic data. Unlike traditional seismic interpretation tools that just “check the box” for AI through black box algorithms, InteractivAI puts the interpreter in the driver’s seat by presenting an intuitive, live feedback loop. Users experience a faster and more comprehensive interpretation, leading to higher-confidence decision-making. Learn more at Bluware.
Guest Bio
Morten Ofstad has worked with computer graphics since graduating from high school. As one of the first employees of Norwegian games developer Funcom, he created the game engine for the 2D games that formed the basis of Funcom’s initial growth. He’s been working as the lead developer of several successful game titles from studios like Sony Computer Entertainment Europe in London and Innerloop Studios in Oslo. Between jobs in the games industry, he completed an M.Sc. in computer science at the University of Oslo, graduating with honors. Besides 3D graphics, Morten’s interests include compiler technology, system architecture, and image processing.
Main Themes
- Shifting from Model-Centric to Data-Centric AI: The podcast highlights the limitations of traditional model-centric AI approaches in seismic interpretation and advocates for a shift towards data-centric approaches.
- Interactive Deep Learning: This approach, championed by Bluware, emphasizes user interaction and control over the training process, allowing for iterative refinement of models and integration of domain expertise.
- Importance of Data Quality and Labeling: The episode stresses the critical role of high-quality data and accurate labeling in training effective AI models for seismic interpretation.
- VDS Data Format: The discussion introduces Bluware’s VDS data format, designed for efficient random access, compression, and streamlined data preparation in AI workflows.
Key Ideas & Facts
- Limitations of Model-Centric AI
- Benefits of Data-Centric & Interactive Deep Learning
- Addressing Challenges of Data Quality and Labeling
- VDS Data Format as an Enabler
Key Concepts and Terms
- Interactive deep learning: A method of AI where the user interacts with the model during training, enabling them to guide the learning process and improve results.
- Model-centric approach: An AI approach focused on optimizing the model’s architecture and parameters while keeping the training data relatively static.
- Data-centric approach: An AI approach that emphasizes the importance of high-quality, accurately labeled data for training, often involving iterative improvement of the dataset alongside model training.
- Black box deep learning: Refers to AI models whose internal workings and decision-making processes are opaque and not easily interpretable by humans.
- VDS (Visualization Data Store): A data format designed for efficient storage and access of seismic data, particularly beneficial for visualization and machine learning tasks.
- OpenVDS: An open-source implementation of the VDS data format, making it accessible for use outside of Bluware’s tools.
- OSDU (Open Subsurface Data Universe): An industry initiative to standardize and facilitate data exchange in the energy sector.
- SEG-Y: A widely used but older standard format for seismic data, often less flexible and efficient than newer alternatives like VDS.
Call to Action
- Explore how data-centric AI tools can be integrated into geoscientists’ workflows.
- Move beyond simply asking questions and receiving answers, and instead utilize AI to “interrogate your data” and gain deeper insights.
Download this Episode
Please visit this page to download this episode as an .MP3 file.
Show Credits
Andrew Geary at TreasureMint hosted, edited, and produced this episode. The SEG podcast team comprises Jennifer Cobb, Kathy Gamble, and Ally McGinnis.
If you have episode ideas or feedback for the show or want to sponsor a future episode, email the show at [email protected].
Historical podcasts are stored in the SEG Basecamp. Please request access to this podcast by emailing [email protected].