🎰 deep learning

1. Can we find the least number of examples possible for training a deep learning network?

2. How can deep learning benefit exploration geophysics?

3. Is deep learning worth the effort over physics-based exploration methods?

📡🎰 enhancement of physics-based exploration methods

This toy problem explores questions No. 1 and No. 2, without trying to substitute physics with deep learning. The code is here.

Imagine we have a physics-based inversion result of the subsurface. Can we make it more accurate?

Let the true subsurface look like Figure 1, and our recovered subsurface with physics-based methods look like Figure 2.

Figure 3 shows the enhanced subsurface results.

We only used 24 training subsurface models. Can we use this learning machine on other subsurface models?

Figure 4 shows subsurface models that were not used during training. Figure 5 shows the recovered models from the learning machine.