Supervised Learning is learning by example. It requires a training set with examples of input data tied to known output. The process of linking input to output is called "labeling".

In OpendTect, we support four key labeling methods for applications involving seismic and well data:

  • Modeling: Creates unlimited examples, but modelling is difficult and is typically only done by experts
  • Processing: Simplest method, requires having input & processed output
  • Interpretation of Points: Relatively easy to interpret, used for detection of sub-seismic and fuzzy objects, but application of image-to-point models is time-consuming
  • Interpretation of Objects: For macro-scale objects, image-to-image models are heavy on computer resources during training but very fast in the application phase and can often be re-used on unseen data.

The video gives more details about each labeling method and guides you to choose the right Machine Learning model for a given task.