Na próxima semana, o GPAM (Grupo de Pesquisa em Aprendizado de Máquina) receberá a visita técnica do Dr. Leander Schietgat, da Universidade Católica de Leuven, Bélgica, como parte das atividades previstas nos projetos de pesquisa financiados pela Universidade Católica de Leuven (Bélgica), CNPq e Fapesp.
http://www.bv.fapesp.br/pt/auxilios/94199/hiper-heuristicas-e-arvores-de-decisao-para-problemas-de-classificacao-hierarquica-e-multi-rotulo/
Aproveitando a oportunidade, gostaríamos de convidá-los para um seminário - na área de Inteligência Computacional - que será apresentado pelo Dr. Leander Schietgat, que também falará sobre oportunidades de colaborações com a KU Leuven.
Apresentador: Dr. Leander Schietgat, Declarative Languages and Artificial Intelligence (DTAI) Group, Department of Computer Science, KU Leuven, Belgium
Data: 18/10/2016 (3a-feira)
Horário: 17:45h
Local: Auditório do Parque Tecnológico (4o andar) - ICT-UNIFESP
Título: Decision trees for multi-target prediction
Resumo:
Multi-target prediction, which is attracting a lot of interest in the machine learning community, is a variant of prediction where instances have multiple target variables. It can take many different forms: for example, in hierarchical multi-label classification, instances may belong to multiple classes at the same time, and these classes are organized in a hierarchy. Applications of multi-target prediction include protein function prediction, scene annotation, the prediction of river water quality parameters, and many more.
There are several approaches to the induction of decision trees for multi-target prediction: learning a single tree (which makes predictions for all targets together) or learn a set of trees (one for each target). Interestingly, the former approach outperforms the latter along three dimensions: predictive accuracy, model size, and induction time. Moreover, ensemble methods of multi-target trees are more accurate than single trees and are competitive with state-of-the-art methods. Many instantiations of multi-target prediction are implemented in Clus, which is a decision tree learning system that works in the predictive clustering trees framework.
In this talk I will give a brief introduction to machine learning, discuss the multi-target prediction setting, and look into a few applications where we use decision trees as learning algorithm.
