Structured and Model-Based Algorithms for Online Optimization
Topic: Optimization | All
Date : 13/02/2023 De 16h00 A 17h00
SPEAKER
Nicola Bastianello, KTH Royal Institute of Technology, Sweden
TITLE
Structured and Model-Based Algorithms for Online Optimization.
ABSTRACT
In many applications ranging from control to image processing to machine learning, we are interested in solving optimization problems that change over time, modeling a dependence on e.g. streaming sources of data. These online optimization problems have attracted much research interest in the last few years, giving rise to an interesting set of challenges and opportunities. In this talk I will discuss the use of structured algorithms for online optimization, which exploit model-based techniques to improve the performance. I will discuss first the class of prediction-correction methods, which leverage a prediction of future problems to achieve better tracking of the optimal solutions trajectory. I will then discuss a robust control-based approach to the design of structured algorithms when a model of the problem's variation over time is available. Numerical results on benchmark problems in signal processing and learning will be presented.
BIO
Nicola Bastianello is a post-doc at the School of Electrical Engineering and Computer Science, KTH Royal Institute of Technology, Sweden under the supervision of Prof. Karl H. Johansson. From 2021 to 2022 he was a post-doc at the Department of Information Engineering (DEI), University of Padova, Italy, where he had receivef the Ph.D. in Information Engineering 2021. During the Ph.D. he was a visiting student at the Department of Electrical, Computer, and Energy Engineering (ECEE), University of Colorado Boulder, Colorado, USA. He received the master degree in Automation Engineering (2018) and the bachelor degree in Information Engineering (2015) from the University of Padova, Italy. His main research interests are online optimization and distributed optimization.
SPEAKER
Nicola Bastianello, KTH Royal Institute of Technology, Sweden
TITLE
Structured and Model-Based Algorithms for Online Optimization.
ABSTRACT
In many applications ranging from control to image processing to machine learning, we are interested in solving optimization problems that change over time, modeling a dependence on e.g. streaming sources of data. These online optimization problems have attracted much research interest in the last few years, giving rise to an interesting set of challenges and opportunities. In this talk I will discuss the use of structured algorithms for online optimization, which exploit model-based techniques to improve the performance. I will discuss first the class of prediction-correction methods, which leverage a prediction of future problems to achieve better tracking of the optimal solutions trajectory. I will then discuss a robust control-based approach to the design of structured algorithms when a model of the problem's variation over time is available. Numerical results on benchmark problems in signal processing and learning will be presented.
BIO
Nicola Bastianello is a post-doc at the School of Electrical Engineering and Computer Science, KTH Royal Institute of Technology, Sweden under the supervision of Prof. Karl H. Johansson. From 2021 to 2022 he was a post-doc at the Department of Information Engineering (DEI), University of Padova, Italy, where he had receivef the Ph.D. in Information Engineering 2021. During the Ph.D. he was a visiting student at the Department of Electrical, Computer, and Energy Engineering (ECEE), University of Colorado Boulder, Colorado, USA. He received the master degree in Automation Engineering (2018) and the bachelor degree in Information Engineering (2015) from the University of Padova, Italy. His main research interests are online optimization and distributed optimization.