Self-supervised active learning for the inversion of geophysical data
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
A novel, self-supervised approach to machine learning inversion for geophysical data is developed. It leverages a combination of supervised, unsupervised, reinforcement, and active learning paradigms working in an iterative scheme. Active learning is used to select the most informative samples from a data pool using heuristics based on sample diversity and uncertainty. Unsupervised learning is used to identify data classes and select the most diverse and informative data. Reinforcement learning uses physics as an auxiliary task, to label the selected samples, which are then utilized in supervised learning. Iterative data augmentation is implemented to steer the training data distribution towards the field data distribution. The machine learning performance is enhanced, allowing for selfsupervised learning. This approach offers several benefits, including sharper results and a more comprehensive exploration of the parameter space with a parsimonious utilization of the computer resources. The method is demonstrated on synthetic data and to exploration-scale geophysical surveys, such as helicopter-borne transient electromagnetics and near surface seismic full waveform inversion in complex geological settings.
Fields of Study & Subjects
Geophysical and Geoelectrical Methods
Seismic Imaging and Inversion Techniques
Geophysical Methods and Applications
Publication details
- Journal / source
- Second International Meeting for Applied Geoscience & Energy
- Volume
- 1
- Pages
- 1025–1029
- Publication date
- 10 February 2026
- Language
- EN