EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
2662. Variational Auto-Encoders and Generative Adversarial Networks for scenario generation
Invited abstract in session MC-35: Urban Logistics and sustainable TRAnsportation: OPtimization under uncertainTY and MAchine Learning, stream Stochastic, Robust and Distributionally Robust Optimization.
Monday, 12:30-14:00Room: 44 (building: 303A)
Authors (first author is the speaker)
| 1. | Enza Messina
|
| DISCo - Department of Informatics, Systems and Communication, University of Milano Bicocca | |
| 2. | Michele Carbonera
|
| Università Milano Bicocca | |
| 3. | Michele Ciavotta
|
| University of Milano Bicocca |
Abstract
In this talk, we address the problem of learning multivariate distribution from empirical data aimed at scenario generation. Capturing the complex spatio-temporal relationship among multiple variables is a challenging task. Copula based models are usually applied in this context for their ability to separate the multivariate structure from its marginal distributions. However, when considering real case studies choosing the right family of copula models may be difficult.
To overcome these difficulties, in this work, we propose a data-driven (or model-free) approach by adopting generative methods. In particular, we explore the use of Variational Auto-encordes and Generative adversarial Networks, for learning the multi-variate joint probability distribution of link speeds on a road network, using real sensor data.
Experimental results, conducted on three distinct benchmark datasets, highlight the potential of the proposed model in generating new scenario samples of multivariate variables hat preserve correlations among variables, while producing samples that faithfully represent the empirical marginal distributions.
Keywords
- Machine Learning
- Analytics and Data Science
- Programming, Stochastic
Status: accepted
Back to the list of papers