If there is one thing this Deep Learning community can all get behind, it is 4D in open source.
For 3D, we have the tools - convolutional neural network CNN for image recognition (seismic sections), recurrent neural network RNN, or specifically long-short term memory LSTM (time series like decline curve), multivariate regression for geospatial analytics. Many recently published papers show practical applications of adapting open-sourced ML models.
To top off using repeated seismic and calibrated production data on a scheduled basis, it is not too early to contemplate the application of reinforcement learning, e.g., generative adversarial network GAN (game like approach using generative model and totally unsupervised learning, enabling machines to beat the Chess and Go world champions).
Once marketing got ahead of what 4D technology could deliver, “if you can’t see it in 3D, it will just pop up in 4D.” It took a decade to master repeatability in acquisition (recording at the same position) and processing (calibrating amplitudes and wavelets) to deliver the benefits of time-lapsed seismic.
Q: Will 4D going open-source accelerate the pace of innovation?
Feel free to chime in.