While studying Bayesian Learning, I have encountered the term, Aleatoric and Epistemic uncertainty, but I have just found it a bit confusing to understand. I believe I haven't found good references to understand the term. What do they mean and what could be examples for these two terms? Hope some plain literal explanation.
$\begingroup$
$\endgroup$
3
asked Oct 30, 2020 at 13:49
-
2$\begingroup$ This is the rare case (concerning statistical terminology) where consulting an English dictionary will actually help clear things up. $\endgroup$– whuber ♦Commented Oct 30, 2020 at 16:35
-
$\begingroup$ Also a rare case of a statistical text incorporating terminology from the humanities and, more specifically, critical theory. See, e.g., The Cultural Studies Reader amazon.com/Cultural-Studies-Reader-Simon-During/dp/0415374138/… $\endgroup$– user234562Commented Oct 30, 2020 at 17:38
-
$\begingroup$ See also stats.stackexchange.com/questions/332026/… $\endgroup$– kjetil b halvorsen ♦Commented Apr 5, 2023 at 16:15
Add a comment
|
2 Answers
$\begingroup$
$\endgroup$
2
A short and very simplified literal explanation:
Aleatoric: uncertainty about the result of an experiment that we can repeat, e.g. dice roll. What is the probability of rolling a 6? - the view of frequency statistics
Epistemic: uncertainty stemming from insufficient knowledge, e.g. one-time experiment (no repeating). What is the probability that - as a result of global warming - the average temperature will be 2 degrees higher in 2050?
I hope it helped :-)
-
$\begingroup$ Is Aleatoric for Frequentist and is Epistemic for Bayesian? $\endgroup$ Commented Nov 2, 2020 at 18:00
-
$\begingroup$ In general, I think yes. The epistemic is at most the view of Bayesian statistics, but there is not a strict wall between them. You can use Bayesian approach for frequency problems too. Let we have an apriori distribution and we can make some experiments and then based on these experiments (and taking into consideration the apriori distribution) we get the posterior distribution. :) $\endgroup$– jumpiniCommented Nov 15, 2020 at 11:49
$\begingroup$
$\endgroup$
Aleatoric Uncertainty: This is the uncertainty of the process which you are trying to model. Say, you want to train a model with some sensor output where the sensor is itself producing some random fluctuations in its reading. So, no matter how many data points you take, your model will not be able to reduce that uncertainty, as the data itself has uncertainty in it. So, aleatoric uncertainty is the uncertainty in data.
Epistemic Uncertainty: This uncertainty happens for lack of enough data to estimate the model parameters properly. If you add more data, this uncertainty will be reduced. So, epistemic uncertainty is the uncertainty in model parameters.
