#13: Scaling ML from a small dataset to complete project

Seismic methods are getting more sophisticated trying to improve subsurface imaging and get more detailed and accurate information about oil and gas prospects. This leads to massively increased size of acquired seismic volumes and needed computational power to process and interpret the data.What can we do to overcome new limitation? Instead of trying to interpret every single piece of data and spend ages to do it, we can expand our knowledge from a tiny piece of data to the whole volume! Unfortunately, conventional seismic interpretation techniques can’t do this. But Deep Learning can!

Using 105 Gb Poseidon seismic data from N-W Australia we have scaled Deep Learning fault interpretation from a small part of the data to the whole dataset.