#mang #dso

aleatory uncertainteies are objective
stochastic in nature
cant do anything aobut them
can be modelled as a pdf
temperature, material params

epistemic reflect the designers lack of knowledge
no pdf
uncertainty of numerical methods


deterministic uncertainties are something like $\pm 2$, defined domain
aleatory and epistemic

probablilistic - likelihood of soemthign ahppening
only aleatory

can convert from epistemic to aleatory if you do a bunch of measurements


# ways to handle this
## robust regularisation
you design for $x$ as well as $x \pm \varepsilon$
not very good 
basically a safety factor
$\min \max f(x\pm \varepsilon)$


$$F_1(x) = \int f(x+\varepsilon) P(\varepsilon) d\varepsilon$$
$$F_2(2) = \int (f(x + \varepsilon) -f(x))^2 p(\varepsilon) d\varepsilon$$
$F_1$ is kinda the expected value
$F_2$ is kinda how flat it is