Uncertainty analysis for neural networks
A method includes receiving geophysical data representative of a geophysical structure; providing the geophysical data as one or more input data to a neural network; training the neural network to reconstruct the geophysical structure that was received and provide one or more uncertainty metrics for one or more features of the geophysical structure that is reconstructed; reconstructing, using the neural network that has been trained, the geophysical structure; and determining, using the neural network that has been trained, the one or more uncertainty metrics by implementing a second drop out condition on the one or more nodes of the one or more hidden layers of the neural network. The training is performed at least partially by implementing a first drop out condition on one or more nodes of one or more hidden layers of the neural network to randomly set an output of the one or more nodes to zero.
1. A method comprising:
receiving interpretation data to analyze for uncertainty, wherein the interpretation data corresponds to a machine learning output representative of a geophysical structure;
conducting a training phase to train a deep learning neural network using the interpretation data, wherein the training phase comprises:
modifying the interpretation data via a first drop out condition applied to one or more nodes of one or more hidden layers of the deep learning neural network to randomly set an output of the one or more nodes to zero;
providing the modified interpretation data as input data to train the deep learning neural network; and
training the deep learning neural network using the modified interpretation data to generate a trained deep learning neural network;
subsequent to the training phase, conducting a prediction phase to reproduce the geophysical structure utilizing the trained deep learning neural network, wherein the prediction phase comprises:
enabling a second drop out condition on the one or more nodes of the one or more hidden layers of the trained deep learning neural network;
generating an output, using the trained deep learning neural network, wherein the generated output comprises a reconstructed geophysical structure; and
determining, using the trained deep learning neural network, one or more uncertainty metrics associated with one or more features of the reconstructed geophysical structure, wherein the one or more uncertainty metrics correspond to an accuracy of the reconstructed geophysical structure and are indicative of a reliability of the interpretation data, and wherein the one or more uncertainty metrics comprise an entropy for classification modeling and one or more distribution moments for regression modeling;
outputting the reconstructed geophysical structure and the one or more uncertainty metrics for display on a computer display; and
displaying, on the computer display, the reconstructed geophysical structure and the one or more uncertainty metrics, wherein the one or more uncertainty metrics are displayed using graphical visual icons.
2. The method of claim 1 , wherein the interpretation data comprises at least one of poststack seismic amplitude data, prestack seismic amplitude data, seismic amplitude data, seismic amplitude derived properties, well log data, well log derived properties, structure interpretation data, stratigraphy interpretation data, or lithology interpretation data.
3. The method of claim 2 , wherein the well log data comprises at least one of gamma ray (GR) logs, neutron porosity (NPHI) logs, or bulk density (RHOZ) logs.
4. The method of claim 1 , wherein the deep learning neural network, the trained deep learning neural network, or both comprises a convolutional neural network (CNN), a pointwise autoencoder with about 3 to about 20 dense layers, a long short-term memory (LSTM) network, a bidirectional LSTM sequential autoencoder, or a 2D/1D-CNN autoencoder with about 5 to about 20 convolutional blocks.
5. The method of claim 1 , wherein the one or more hidden layers of the deep learning neural network and the trained deep learning neural network comprises about 20 to about 200 hidden layers, about 50 to about 150 hidden layers, or about 75 to about 125 hidden layers.
6. The method of claim 1 , wherein the first dropout condition is based on a Bernoulli distribution, where the Bernoulli condition sets an output of the one or more nodes of the one or more hidden layers to zero.
7. The method of claim 1 , wherein modifying the interpretation data comprises performing a regularization process to reduce a complexity of the input data provided to the deep learning neural network.
8. The method of claim 1 , further comprising:
repeatedly conducting the prediction phase to reproduce the geophysical structure by enabling one or more additional drop out conditions using the input data to generate one or more additional reconstructed geophysical structures, wherein the input data is identical from a previous iteration of the deep learning neural network and the trained deep learning neural network with the first drop out condition or a different drop out condition; and
determining one or more additional uncertainty metrics based on the one or more additional reconstructed geophysical structures that are repeatedly generated.
9. The method of claim 8 , further comprising quantifying the one or more uncertainty metrics and the one or more additional uncertainty metrics based on an entropy of aggregated results of the reconstructed geophysical structure and the one or more additional reconstructed geophysical structures.
10. The method of claim 9 , wherein the aggregated results are determined based on a confidence interval.
11. The method of claim 1 , wherein a type of the one or more uncertainty metrics is based on a type of the geophysical structure being modeled.
12. A computing system, comprising:
one or more processors; and
a memory system comprising one or more non-transitory, computer-readable media storing instructions that, when executed by at least one of the one or more processors, cause the computing system to perform operations, the operations comprising:
receiving interpretation data to analyze for uncertainty, wherein the interpretation data corresponds to a machine learning output representative of a geophysical structure;
conducting a training phase to train a deep learning neural network using the interpretation data, wherein the training phase comprises:
modifying the interpretation data via a first drop out condition applied to one or more nodes of one or more hidden layers of the deep learning neural network to randomly set an output of the one or more nodes to zero;
providing the modified interpretation data as input data to train the deep learning neural network; and
training the deep learning neural network using the modified interpretation data to generate a trained deep learning neural network;
subsequent to the training phase, conducting a prediction phase to reproduce the geophysical structure utilizing the trained deep learning neural network, wherein the prediction phase comprises:
enabling a second drop out condition on the one or more nodes of the one or more hidden layers of the trained deep learning neural network;
generating an output, using the trained deep learning neural network, wherein the generated output comprises a reconstructed geophysical structure; and
determining, using the trained deep learning neural network, one or more uncertainty metrics associated with one or more features of the reconstructed geophysical structure, wherein the one or more uncertainty metrics correspond to an accuracy of the reconstructed geophysical structure and are indicative of a reliability of the interpretation data, and wherein the one or more uncertainty metrics comprise an entropy for classification modeling and one or more distribution moments for regression modeling;
outputting the reconstructed geophysical structure and the one or more uncertainty metrics for display on a computer display; and
displaying, on the computer display, the reconstructed geophysical structure and the one or more uncertainty metrics, wherein the one or more uncertainty metrics are displayed using graphical visual icons.
13. The computing system of claim 12 , wherein the interpretation data comprises at least one of poststack seismic amplitude data, prestack seismic amplitude data, seismic amplitude data, seismic amplitude derived properties, well log data, well log derived properties, structure interpretation data, stratigraphy interpretation data, or lithology interpretation data.
14. The computing system of claim 12 , wherein the deep learning neural network, the trained deep learning neural network, or both comprise a convolutional neural network (CNN), a pointwise autoencoder with about 3 to about 20 dense layers, a long short-term memory (LSTM) network, a bidirectional LSTM sequential autoencoder, or a 2D/1D-CNN autoencoder with about 5 to about 20 convolutional blocks.
15. The computing system of claim 12 , wherein the one or more hidden layers of the deep learning neural network and the trained deep learning neural network comprises about 20 to about 200 hidden layers, about 50 to about 150 hidden layers, or about 75 to about 125 hidden layers.
16. The computing system of claim 12 , wherein the first dropout condition is based on a Bernoulli distribution, where the Bernoulli condition sets an output of the one or more nodes of the one or more hidden layers to zero.
17. The computing system of claim 12 , wherein the operations further comprise:
repeatedly conducting the prediction phase to reproduce the geophysical structure by enabling one or more additional drop out conditions using the input data to generate one or more additional reconstructed geophysical structures, wherein the input data is identical from a previous iteration of the deep learning neural network and the trained deep learning neural network with the first drop out condition or a different drop out condition; and
determining one or more additional uncertainty metrics based on the one or more additional reconstructed geophysical structures that are repeatedly generated.
18. A non-transitory, computer-readable medium storing instructions that, when executed by at least one processor of a computing system, cause the computing system to perform operations, the operations comprising:
receiving interpretation data to analyze for uncertainty, wherein the interpretation data corresponds to a machine learning output representative of a geophysical structure;
conducting a training phase to train a deep learning neural network using the interpretation data, wherein the training phase comprises:
modifying the interpretation via a first drop out condition applied to one or more nodes of one or more hidden layers of the deep learning neural network to randomly set an output of the one or more nodes to zero;
providing the modified interpretation data as input data to train the deep learning neural network; and
training the deep learning neural network using the modified interpretation data to generate a trained deep learning neural network;
subsequent to the training phase, conducting a prediction phase to reproduce the geophysical structure utilizing the trained deep learning neural network, wherein the prediction phase comprises:
enabling a second drop out condition on the one or more nodes of the one or more hidden layers of the trained deep learning neural network;
generating an output, using the trained deep learning neural network, wherein the generated output comprises a reconstructed geophysical structure;
determining, using the trained deep learning neural network, one or more uncertainty metrics associated with one or more features of the reconstructed geophysical structure, wherein the one or more uncertainty metrics correspond to an accuracy of the reconstructed geophysical structure and are indicative of a reliability of the interpretation data, and wherein the one or more uncertainty metrics comprise an entropy for classification modeling and one or more distribution moments for regression modeling;
outputting the reconstructed geophysical structure and the one or more uncertainty metrics for display on a computer display; and
displaying, on the computer display, the reconstructed geophysical structure and the one or more uncertainty metrics, wherein the one or more uncertainty metrics are displayed using graphical visual representations.