EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
620. Defining and Quantifying the Marginal Value of a Data Set via Distributionally Robust Optimization
Invited abstract in session WA-35: Data Valuation from Data-driven Optimization, stream Stochastic, Robust and Distributionally Robust Optimization.
Wednesday, 8:30-10:00Room: 44 (building: 303A)
Authors (first author is the speaker)
| 1. | Juan Miguel Morales
|
| Applied Mathematics, University of Málaga | |
| 2. | Robert Mieth
|
| Industrial and Systems Engineering, Rutgers University | |
| 3. | Adrián Esteban-Pérez
|
| Rotterdam School of Management, Erasmus University of Rotterdam |
Abstract
Data-driven stochastic optimization has empowered decision-making by providing an effective way to exploit data to tackle complex problems in many domains. However, there is still a lack of theory and methods to quantify the sensitivity of the decision cost to the natural input to stochastic optimization, namely, the data set. In this talk we show that Distributionally Robust Optimization (DRO) offers a natural framework to perform this sensitivity analysis by establishing the notion of marginal value of the quality of a data set. We discuss the ability of the Wasserstein metric to encode data quality and then introduce a Wasserstein DRO formulation to compute the marginal value of data sets from multiple sources or providers that may refer to the same uncertain input parameter and/or to different ones.
Keywords
- Programming, Stochastic
- Decision Support Systems
- Stochastic Optimization
Status: accepted
Back to the list of papers