On Learning for Ambiguous Chance Constrained Problems
Authors:
A Ch Madhusudanarao,
Rahul Singh
Abstract:
We study chance constrained optimization problems $\min_x f(x)$ s.t. $P(\left\{ θ: g(x,θ)\le 0 \right\})\ge 1-ε$ where $ε\in (0,1)$ is the violation probability, when the distribution $P$ is not known to the decision maker (DM). When the DM has access to a set of distributions $\mathcal{U}$ such that $P$ is contained in $\mathcal{U}$, then the problem is known as the ambiguous chance-constrained p…
▽ More
We study chance constrained optimization problems $\min_x f(x)$ s.t. $P(\left\{ θ: g(x,θ)\le 0 \right\})\ge 1-ε$ where $ε\in (0,1)$ is the violation probability, when the distribution $P$ is not known to the decision maker (DM). When the DM has access to a set of distributions $\mathcal{U}$ such that $P$ is contained in $\mathcal{U}$, then the problem is known as the ambiguous chance-constrained problem \cite{erdougan2006ambiguous}. We study ambiguous chance-constrained problem for the case when $\mathcal{U}$ is of the form $\left\{μ:\frac{μ(y)}{ν(y)}\leq C, \forall y\inΘ, μ(y)\ge 0\right\}$, where $ν$ is a ``reference distribution.'' We show that in this case the original problem can be ``well-approximated'' by a sampled problem in which $N$ i.i.d. samples of $θ$ are drawn from $ν$, and the original constraint is replaced with $g(x,θ_i)\le 0,~i=1,2,\ldots,N$. We also derive the sample complexity associated with this approximation, i.e., for $ε,δ>0$ the number of samples which must be drawn from $ν$ so that with a probability greater than $1-δ$ (over the randomness of $ν$), the solution obtained by solving the sampled program yields an $ε$-feasible solution for the original chance constrained problem.
△ Less
Submitted 11 February, 2024; v1 submitted 31 December, 2023;
originally announced January 2024.
