On the Optimal Bias-Variance Trade-off in High Dimensions
Event details
| Date | 15.05.2023 |
| Hour | 11:00 › 12:00 |
| Speaker | Riccardo Cescon, University of Padova, Italy |
| Location | |
| Category | Conferences - Seminars |
| Event Language | English |
Abstract:
In this talk we will concentrate on the classical bias-variance trade-off in linear regression, for problems whose dimension and number of samples are very large. Such problems present many challenges, such as the high-dimensionality and the nonconvexity of the variance regularizer. We are interested in studying the performance of the estimators and in understanding how to optimally choose the variance regularization parameter. We will show that, despite the aforementioned challenges, such problems can be tackled using tools from distributionally robust optimization and high-dimensional statistics.
Bio:
Riccardo Cescon holds a bachelor’s degree in Information Engineering (2020) and a master’s degree in Control Systems Engineering (2022) both with honors from the University of Padova, Italy. During his master’s thesis he spent six months at ETHZ as a visiting student at the Institut für Automatik under the supervision of prof. Florian Dörfler working on the optimal bias-variance tradeoff in high dimensions. He’s currently a research assistant at the Department of Information Engineering at the University of Padova under the supervision of prof. Ruggero Carli on the European project Drapebot.
His research interests include control theory, machine learning and optimization.
In this talk we will concentrate on the classical bias-variance trade-off in linear regression, for problems whose dimension and number of samples are very large. Such problems present many challenges, such as the high-dimensionality and the nonconvexity of the variance regularizer. We are interested in studying the performance of the estimators and in understanding how to optimally choose the variance regularization parameter. We will show that, despite the aforementioned challenges, such problems can be tackled using tools from distributionally robust optimization and high-dimensional statistics.
Bio:
Riccardo Cescon holds a bachelor’s degree in Information Engineering (2020) and a master’s degree in Control Systems Engineering (2022) both with honors from the University of Padova, Italy. During his master’s thesis he spent six months at ETHZ as a visiting student at the Institut für Automatik under the supervision of prof. Florian Dörfler working on the optimal bias-variance tradeoff in high dimensions. He’s currently a research assistant at the Department of Information Engineering at the University of Padova under the supervision of prof. Ruggero Carli on the European project Drapebot.
His research interests include control theory, machine learning and optimization.
Practical information
- General public
- Free