10.6084/m9.figshare.7422629.v1 Jesper Soeren Dramsch Jesper Soeren Dramsch Mikael Lüthje Mikael Lüthje Deep Learning: From Cats to 4D Seismic - Reducing cycle time and model training cost in asset management figshare 2018 geophysics 4D seismic time-lapse seismic deep learning transfer learning Seismology and Seismic Exploration Geophysics 2018-12-05 06:01:15 Poster https://figshare.com/articles/poster/Deep_Learning_From_Cats_to_4D_Seismic_-_Reducing_cycle_time_and_model_training_cost_in_asset_management/7422629 4D Seismic data has proven invaluable in O&G asset management, however, it’s engineering challenges are still plentiful. These challenges include non-repeatable noise, tie-in and match with production curves, as well as, separation of imaging, pressure and saturation effects. Deep learning has proven robust at separating effects [1] with a strong data-dependent prior and has been shown effective in modelling physics-based systems [2]. We present work that reduces training times and thus reduces cost of implementation and enables rapid prototyping of<br>experiments. This can be used in seismic modelling, physical effect separation, time series alignment and automatic seismic interpretation. <br>