Learning and Optimization

• "Learning for Spatial Branching: An Algorithm Selection Approach", organizado polo CITMAga en formato presencial. Será impartido por Brais González Rodríguez (Universidade de Santiago de Compostela)

Data: xoves 1 de decembro

Hora: 15:30 h.

Lugar: Aula seminario departamento Estatística e Investigación Operativa (USC)

Abstract: 

The use of machine learning techniques to improve the performance of branch-and-bound optimization algorithms is a very active area in the context of mixed integer linear problems, but little has been done for non-linear optimization. To bridge this gap, we develop a learning framework for spatial branching and show its efficacy in the context of the Reformulation-Linearization Technique for polynomial optimization problems. The proposed learning is performed offline, based on instance-specific features and with no computational overhead when solving new instances. Novel graph-based features are introduced, which turn out to play an important role for the learning. Experiments on different benchmark instances from the literature show that the learning-based branching rule significantly outperforms the standard rules.

 

• "Alternating projection approach for a continuous formulation of convex and cardinality-constrained optimization", organizado polo CITMAga en formato presencial. Será impartido por Marcos Raydan (NOVAMATH-CMA, Universidade NOVA de Lisboa)

Data: xoves 1 de decembro

Hora: 16:00 h.

Lugar: Aula seminario departamento Estatística e Investigación Operativa (USC)

Abstract:

Taking advantage of a recently developed continuous formulation that relaxes the cardinality constraint, we propose a specialized penalty gradient projection scheme combined with alternating projection ideas to solve these problems.  We focus on the standard mean-variance portfolio optimization problem for which we can only invest in a preestablished limited number of assets. The approach can be adapted to find the sparsest possible solution of a constrained underdetermined linear system of equations.

 

• "Counterfactual Decisions: A Mathematical Optimization Problem in Explainable Machine Learning", organizado polo CITMAga en formato presencial. Será impartido por Emilio Carrizosa (IMUS, Universidad de Sevilla)

Data: xoves 1 de decembro

Hora: 16:30 h.

Lugar: Aula seminario departamento Estatística e Investigación Operativa (USC)

Abstract:

Machine Learning-based classification models are frequently criticized because of their lack of transparency (black boxes).  To remedy this, a variety of tools are proposed within the field of Explainable Machine Learning. In counterfactual analysis, given a classifier and a record, one seeks the minimal (feasible!) perturbation needed in its attributes to make the classifier change the classification. The problem of finding counterfactuals for one or several records yields interesting mathematical optimization problems, which will be revised in this talk.

 

Actividade co-financiada coa colaboración da Consellería de Cultura, Educación, Formación Profesional e Universidades