Exploring the RCNN Technique in a Multiple-Point Statistics Framework
Loading...
Date
2019
Authors
Avalos, Sebastian
Ortiz, Julian M.
Keyword
Abstract
This work explores the Recursive Convolutional Neural Network (RCNN) technique in terms of describing (1) the inner architecture, (2) the training process and (3) the simulation algorithm in a multiple-point statistics simulation framework. To acquire a more intuitive understanding of the previous description, visualizations of hidden layers activations using three different training images is presented. Two interesting applications (1) using RCNN E-types as secondary information for MPS algorithms and (2) training RCNN with scarce and limited information, are also explored.