Universidade Federal do Rio de Janeiro COPPE

Instituto Alberto Luiz Coimbra de Pós-Graduação e Pesquisa de Engenharia

Instituto de Matemática

 
Visualizar Meses
Visualizar Meses
Visualizar Flat
Visualizar Flat
Visualizar Semanas
Visualizar Semanas
Visualizar Dias
Visualizar Dias
Categorias
Categorias
Procurar
Procurar

Evento: 'Seminário Wouter Caarls (DCC/UFRJ), no PESC'

Eventos PESC (Palestras, Seminários, etc.)
Palestras, Seminários, etc. do PESC/COPPE/UFRJ.
Data: Wednesday, September 17, 2014 At 10:00
Duração: 2 Horas

O pesquisador Wouter Caarls está realizando um pós-doutorado supervisionado pelo Prof. Daniel Menasche (DCC/UFRJ) tendo sido agraciado com uma Bolsa Jovens Talentos do programa Ciência sem Fronterias. Sua área de estudo e pesquisa é Aprendizagem por Reforço tema novamente em moda alimentado pelo furacão do Big Data e Data Science. Ele irá proferir uma palestra abrangente no PESC.

Dia 17/09 (quarta) - 10h - sala H-324B

---

Título:
Reinforcement Learning for Robotics

Resumo:
Reinforcement learning is a technique to solve Markov Decision Processes without the need for an explicit transition model. Through trial and error, it finds a control policy that maximizes the received reward. One application area is that of robotics, where it is often easier to specify the relevant task through rewards, rather than through explicitly designed control policies. However, the trial and error nature of the learning process can cause damage to the mechanical system. I will discuss solutions to this problem from two directions: one, to reduce the damage caused by exploratory actions, and two, to reduce the learning time itself through parallelization. As a result, we can perform real-time reinforcement learning on real robotic systems.

Short Bio:
Wouter Caarls received his M.Sc. degree in artificial intelligence from the University of Amsterdam, Amsterdam, The Netherlands. He obtained a Ph.D. from the Delft University of Technology, Delft, The Netherlands on the subject of the automatic optimization of a parallel computer architecture for smart cameras. He currently holds a postdoc position at the Department of Computer Science, Federal University of Rio de Janeiro, Brazil, researching the applications of reinforcement learning in robotics and other areas, specifically through the use of parallelization. His research interests include reinforcement learning, optimization, parallel algorithms, and robot middleware.


Procurar no Calendário

Powered by ExtCalendar 2

© 2017 PESC/COPPE - Programa de Engenharia de Sistemas e Computação

Cidade Universitária, Centro de Tecnologia, Bloco H, Sala 319
Caixa Postal: 68511 CEP: 21941-972 Fones: +55 21 3938-8672 / +55 21 3938-8673 Fax: +55 21 3938-8676
Rio de Janeiro - RJ - Brasil
Horário de atendimento da Secretaria: 2a. a 6a. de 7:00 às 16:00 horas (exceto feriados escolares)