This course focuses on applying Conformal Prediction to probabilistically quantify subsurface uncertainty. It transforms a single prediction from a ML model into a range of possible outcomes, known as a prediction set or interval. This approach allows us to assess how confident we can be in a prediction, rather than just providing a point estimate. Participants will learn to quantify uncertainty in subsurface data through hands-on applications. The course aims to equip geoscientists with the skills to generate reliable probabilistic prediction intervals, improving reservoir characterization and exploration strategies. It combines theoretical knowledge with practical implementation using real-world datasets.
Duration
8 Hours
Prerequisites (Knowledge/Experience/Education Required)
A basic understanding of a programming language (preferably Python) Participants need to have access to a computer with internet access and a Google account, consisting of Gmail and Google Drive as the course uses the Google Colab platform.
Learner Outcomes
Upon completion of this course, learners will be able to:
- Communicate the uncertainty range(s) associated with our predictions and therefore, risks to peers & management.
- Explain what Conformal Prediction is and how it can be used to quantify subsurface uncertainty.
- Understand that Conformal Prediction can be applied to regression, classification and even time series data.
- Apply Conformal Prediction (CP) regression techniques to estimate prediction intervals (for example, P95-P50-P5)
- Apply Conformal Prediction to quantify facies uncertainty conveyed by classification being in sets of potential classes rather than a single prediction
