(for Petroleum Industry)
Speaker: Pavel Temirchev
4th year Ph.D. student
"Cat"
"Cat"
"Dog"
"Giraffe"
Object
Target variable
\(x\)
\(y\)
you need a dataset of examples:
\(\mathcal{D}=\{x_i, y_i\}_i \)
Assume, you have an ML model (e.g., model that mimics tNavigator):
And you collect your dataset as following:
some useful formulas
The entropy:
(the measure of uncertainty)
The Kulback-Leibler divergence:
(kind of distance between two random variables)
(not a metric!)
The Bayes' rule:
(high uncertainty not necessarily signify unobserved data)
(any ideas on how to embed uncertainty into the model?)
One way: train a NN to predict parameters of a distribution. E.g.:
- a NN with two outputs
Use the maximum likelihood approach to train it
with stochastic Neural Networks
Is the petroleum engineering deterministic?
can help us
Now the parameters \(\theta\) are also random variables.
We need to find the posterior given the dataset and a prior:
Now the parameters \(\theta\) are also random variables.
We need to find the posterior given the dataset and a prior:
And the prediction of the BNN:
\(p(\mathcal{D}|\theta)\) - easy
\(p(\theta)\) - is given
\(p(\mathcal{D}) = \int p(\mathcal{D}|\theta) p(\theta) d\theta \) - very hard
\(\theta_i \in n \) trained SNN
it is just an ensemble of NNs
The total uncertainty is the entropy of the BNN's predictive distribution:
The aleatoric uncertainty is the average entropy of the ensemble members:
The epistemic uncertainty is the difference between them:
It shows the difference between ensemble members
We have proposed a possible solution to the dataset acquisition problem for the ROM training: