Defesa de Exame de Qualificação – Alireza Olama – 5/8/2020
30/07/2020 22:48
| Defesa de Exame de Qualificação | |
| Aluno | Alireza Olama |
| Orientador | Prof. Eduardo Camponogara, Dr. – DAS/UFSC |
| Coorientador | Prof. Paulo Renato da Costa Mendes, Dr. – ITWM/Alemanha |
| Data
|
5/8/2020 (quarta-feira) – 8h30
Videoconferência |
|
Banca |
Prof. Daniel Ferreira Coutinho, Dr. – DAS/UFSC (presidente)
Prof. Giuseppe Notarstefano, Dr. – Università di Bologna Prof. Marcus Vinicius Soledade Poggi de Aragão, Dr. – INF/PUC-Rio |
| Título | Distributed Cardinality Constrained Mathematical Optimization for Learning and Control |
| Abstract: Cardinality constrained programming (CCP) problems are one of the significant optimization modeling frameworks for solving a broad class of modern real-world learning and control problems. These types of problems can be modeled as nonlinear mixed-integer programs with special structures and are in general NP-hard. Several methods and algorithms have been proposed for solving the CCP problems in the past few years, however, they are mostly centralized methods and can be applied to medium scale CCP problems. This research proposal aims to develop fully decentralized numerical algorithms to solve generalized CCP problems over Cyber-Physical Networks (CPN) where each node cooperatively deals with a portion of the problem in parallel. This proposal proposes a Distributed Primal Outer Approximation (DIPOA) algorithm as an initial strategy to solve convex CCP problems distributedly. Moreover, the DIPOA algorithm is applied to sparse logistic regression problems for which numerical results also are provided. Finally, based on the studied theories and algorithms, I propose to consider and develop a general algorithmic framework for distributed CCP problems with multiple sources of non-convexity and a more generalized distributed set-up. | |
